Indicative is the leading customer analytics platform offering businesses the ability to analyze customer behavior across all digital touchpoints, without requiring involvement from BI specialists or engineers, or even the knowledge of complicated code or SQL

Indicative

Deepfake – a combination of ‘deep learning’ and ‘fake’ – refers to advanced artificial intelligence used towards the creation of synthetic yet highly realistic images, audio, and video hoaxes[1]. This can be done by creating entirely original material, all of which is entirely false, or by taking pre-existing material and swapping audio/ visual elements with that of another person. Either way, the impact of deepfakes on public trust and wider society has been widely discussed, especially since the proliferation of fake news in 2016 political events[2] (Trump in the US and Brexit in the UK). While deepfakes were once a somewhat intangible threat, to be dealt with in the future, they have now evolved to a level of sophistication that presents genuine, immediate threats to many aspects of society. The likes of OpenAI’s release of text-to-video generative AI, Sora[3], catalyze such risk. In this article, we will explore how deepfakes can erode trust within the legal system, affecting court procedures, evidence gathering, and even free speech. A Brief History of Technology and Court Evidence Types Evidence plays a crucial role in any courtroom but is only ever as good as every one’s ability to trust in it. We started with witness testimony and physical artifacts which are now trumped by voice recordings, photographs, and videos such as CCTV. Technology-based evidence creates more concrete arguments: a video of someone committing a murder is a bit more convincing than the murder weapon being found in their possession. Hence, “historically, audio and video evidence are considered [the] gold standard”[3]. But with all types of evidence, there is room for manipulation. An eyewitness could lie or misinterpret something they see. A murder weapon can be planted, and fingerprints can be tampered with. Once technology is introduced, these manipulations scale in complexity. Voice editing software and photo-editing techniques, e.g., photoshop, can cast doubt on the authenticity of audio and visual evidence. Where with traditional dishonesty there is often a physical trail of evidence to follow, technological manipulation requires a higher level of skill to be able to trace it, e.g., inconsistent time stamps, audio splicing, and minor discrepancies indicative of tampering[4]. With the evolution of technology, the legal system has therefore had to adapt its best practice procedures over time. With regards to evidence gathering and use, this has meant more robust and stringent measures to test authenticity and reliability, as well as a new direction of forensic sourcing/ testing (digital forensics[5]) and chain of custody checks[6]. However, deepfakes now present an even more significant leap in evidence manipulation capabilities. These sophisticated AI-generated media can convincingly depict individuals saying or doing things they never actually did. Unlike previous forms of manipulation, deepfakes go so far as to blur the line between reality and fabrication, posing unprecedented challenges for courts in assessing the authenticity and reliability of audio-visual evidence. Case Studies The most obvious ways deep fakes can be used to help circumvent the law are: The fabrication of evidence to provide alibis for activities and prove innocence.
The fabrication of evidence showing someone’s guilt and involvement in a crime.
Using the possibility of deepfakes to protest the legitimacy of authentic yet incriminating material7. Our first case study looks at fabricating evidence to incriminate someone wrongfully. In 2019, A deepfaked recording was used in a UK custody battle to discredit the father’s worthiness of shared custody. According to the father’s lawyer, Byron James, a “heavily doctored recording” had been presented to the court in which the father is heard making “direct and violent” threats to his wife[8]. However, after further examination, it was found that the recording presented in court had been manipulated to include words that had not been used by the father. In fact, demonstrative of how increasingly accessible editing technologies have become, the mother had used “software and online tutorials to put together a plausible file”[8]. While James had not previously encountered AI-doctored evidence in the courtroom, he has since commented on how it calls into question what kind of evidence can actually be trusted these days. Had they not discovered the tampering, and managed to obtain the original file and metadata, the mother may have successfully persuaded the courtroom as to the father’s fictitiously violent character. While the outcome of, and those involved in the hearing are confidential, it highlights the dangers of taking audio/visual evidence at face value. Not only this, but with the increasingly widespread ease and access to deepfake technologies, versus the relatively older age of judges, the necessary weariness of deep learning technologies may not be fully understood. Our second case study looks at people claiming real videos are deepfakes. In a more high-profile case, Elon Musk’s lawyers attempted to get a lawsuit against Tesla dismissed by claiming Musk had been the victim of deepfake videos. In 2016, Musk was videoed at a tech conference speaking about the extreme safety of Tesla’s Model S and Model X self-driving autonomous features. The videos from this interview have been on YouTube for 7 years now. However, in 2023, the videos resurfaced when a man died after his Tesla crashed while in self-driving mode, and the man’s family and lawyers cited those 2016 claims. Musk’s lawyers tried to deny their authenticity[7]. Being such a public figure, Musk is indeed the subject of deepfakes, which is what his lawyers tried to argue in this instance. The video here was, of course, real though, and the courts did not buy Musk’s lawyers’ claims. However, this highlights how in the age of deepfakes people can not only fabricate but also deny reality. “As people become more aware of how easy it is to fake audio and video, bad actors can weaponize that scepticism”[7]. Especially for public figures, deepfakes offer a layer of protection and immunity behind which they can hide, and avoid taking ownership of reality. Even for non-public figures, instances like this have been seen in courts before, and continue to increase in commonality. Deepfakes and International Regulation Recent regulations targeting deepfakes have emerged globally, with notable initiatives like California's disclosure requirement for AI-generated political ads[9] and the UK's Online Safety Act[10] mandating social media platforms to crack down on harmful content, explicitly including deepfakes. This said, the majority of deepfake regulation currently only really targets deepfakes used with pornographic intent, rather than to circumvent justice systems. While these regulations offer frameworks for accountability and deterrence, their effectiveness hinges on enforcement, resource allocation, skill, and adaptability. Furthermore, although regulation will only grow in necessity for deepfakes under many applications, governments and policymakers will have to consider the potential for negative unintended consequences. This includes impacts on freedom of speech and expression. Vaguely worded laws could lead to censorship or self-censorship, reducing participation in public dialogue and harming democracy. Balancing the objective of combatting deceptive deepfake media with safeguarding human rights will continue as a nuanced challenge, showing the complexities of the ethical considerations surrounding deepfakes, society, and public trust. References 1 What is deepfake AI? A definition from TechTarget 2 Exposure to untrustworthy websites in the 2016 U.S. election - PMC 3 AI deepfakes are poised to enter court proceedings at time of low trust in legal system 4 Authentication of Digital Audio Recordings 5 Understanding Digital Forensics: Process, Techniques, and Tools 6 Digital Forensics and the Chain of Custody: How Is Electronic Evidence Collected and Safeguarded? 7 People are trying to claim real videos are deepfakes. The courts are not amused 8 'Deepfake' audio evidence used in UK court to discredit Dubai dad 9 California Looks to Boost Deepfake Protections Before Elections 10 New laws to better protect victims from abuse of intimate images - GOV.UK Deepfake – a combination of ‘deep learning’ and ‘fake’ – refers to advanced artificial intelligence used towards the creation of synthetic yet highly realistic images, audio, and video hoaxes[1]. This can be done by creating entirely original material, all of which is entirely false, or by taking pre-existing material and swapping audio/ visual elements with that of another person. Either way, the impact of deepfakes on public trust and wider society has been widely discussed, especially since the proliferation of fake news in 2016 political events[2] (Trump in the US and Brexit in the UK). While deepfakes were once a somewhat intangible threat, to be dealt with in the future, they have now evolved to a level of sophistication that presents genuine, immediate threats to many aspects of society. The likes of OpenAI’s release of text-to-video generative AI, Sora[3], catalyze such risk. In this article, we will explore how deepfakes can erode trust within the legal system, affecting court procedures, evidence gathering, and even free speech. A Brief History of Technology and Court Evidence Types Evidence plays a crucial role in any courtroom but is only ever as good as every one’s ability to trust in it. We started with witness testimony and physical artifacts which are now trumped by voice recordings, photographs, and videos such as CCTV. Technology-based evidence creates more concrete arguments: a video of someone committing a murder is a bit more convincing than the murder weapon being found in their possession. Hence, “historically, audio and video evidence are considered [the] gold standard”[3]. But with all types of evidence, there is room for manipulation. An eyewitness could lie or misinterpret something they see. A murder weapon can be planted, and fingerprints can be tampered with. Once technology is introduced, these manipulations scale in complexity. Voice editing software and photo-editing techniques, e.g., photoshop, can cast doubt on the authenticity of audio and visual evidence. Where with traditional dishonesty there is often a physical trail of evidence to follow, technological manipulation requires a higher level of skill to be able to trace it, e.g., inconsistent time stamps, audio splicing, and minor discrepancies indicative of tampering[4]. With the evolution of technology, the legal system has therefore had to adapt its best practice procedures over time. With regards to evidence gathering and use, this has meant more robust and stringent measures to test authenticity and reliability, as well as a new direction of forensic sourcing/ testing (digital forensics[5]) and chain of custody checks[6]. However, deepfakes now present an even more significant leap in evidence manipulation capabilities. These sophisticated AI-generated media can convincingly depict individuals saying or doing things they never actually did. Unlike previous forms of manipulation, deepfakes go so far as to blur the line between reality and fabrication, posing unprecedented challenges for courts in assessing the authenticity and reliability of audio-visual evidence. Case Studies The most obvious ways deep fakes can be used to help circumvent the law are: The fabrication of evidence to provide alibis for activities and prove innocence. The fabrication of evidence showing someone’s guilt and involvement in a crime. Using the possibility of deepfakes to protest the legitimacy of authentic yet incriminating material7. The fabrication of evidence to provide alibis for activities and prove innocence. The fabrication of evidence showing someone’s guilt and involvement in a crime. Using the possibility of deepfakes to protest the legitimacy of authentic yet incriminating material7. Our first case study looks at fabricating evidence to incriminate someone wrongfully. In 2019, A deepfaked recording was used in a UK custody battle to discredit the father’s worthiness of shared custody. According to the father’s lawyer, Byron James, a “heavily doctored recording” had been presented to the court in which the father is heard making “direct and violent” threats to his wife[8]. However, after further examination, it was found that the recording presented in court had been manipulated to include words that had not been used by the father. In fact, demonstrative of how increasingly accessible editing technologies have become, the mother had used “software and online tutorials to put together a plausible file”[8]. While James had not previously encountered AI-doctored evidence in the courtroom, he has since commented on how it calls into question what kind of evidence can actually be trusted these days. Had they not discovered the tampering, and managed to obtain the original file and metadata, the mother may have successfully persuaded the courtroom as to the father’s fictitiously violent character. While the outcome of, and those involved in the hearing are confidential, it highlights the dangers of taking audio/visual evidence at face value. Not only this, but with the increasingly widespread ease and access to deepfake technologies, versus the relatively older age of judges, the necessary weariness of deep learning technologies may not be fully understood. Our second case study looks at people claiming real videos are deepfakes. In a more high-profile case, Elon Musk’s lawyers attempted to get a lawsuit against Tesla dismissed by claiming Musk had been the victim of deepfake videos. In 2016, Musk was videoed at a tech conference speaking about the extreme safety of Tesla’s Model S and Model X self-driving autonomous features. The videos from this interview have been on YouTube for 7 years now. However, in 2023, the videos resurfaced when a man died after his Tesla crashed while in self-driving mode, and the man’s family and lawyers cited those 2016 claims. Musk’s lawyers tried to deny their authenticity[7]. Being such a public figure, Musk is indeed the subject of deepfakes, which is what his lawyers tried to argue in this instance. The video here was, of course, real though, and the courts did not buy Musk’s lawyers’ claims. However, this highlights how in the age of deepfakes people can not only fabricate but also deny reality. “As people become more aware of how easy it is to fake audio and video, bad actors can weaponize that scepticism”[7]. Especially for public figures, deepfakes offer a layer of protection and immunity behind which they can hide, and avoid taking ownership of reality. Even for non-public figures, instances like this have been seen in courts before, and continue to increase in commonality. Deepfakes and International Regulation Recent regulations targeting deepfakes have emerged globally, with notable initiatives like California's disclosure requirement for AI-generated political ads[9] and the UK's Online Safety Act[10] mandating social media platforms to crack down on harmful content, explicitly including deepfakes. This said, the majority of deepfake regulation currently only really targets deepfakes used with pornographic intent, rather than to circumvent justice systems. While these regulations offer frameworks for accountability and deterrence, their effectiveness hinges on enforcement, resource allocation, skill, and adaptability. Furthermore, although regulation will only grow in necessity for deepfakes under many applications, governments and policymakers will have to consider the potential for negative unintended consequences. This includes impacts on freedom of speech and expression. Vaguely worded laws could lead to censorship or self-censorship, reducing participation in public dialogue and harming democracy. Balancing the objective of combatting deceptive deepfake media with safeguarding human rights will continue as a nuanced challenge, showing the complexities of the ethical considerations surrounding deepfakes, society, and public trust. References 1 What is deepfake AI? A definition from TechTarget What is deepfake AI? A definition from TechTarget What is deepfake AI? A definition from TechTarget 2 Exposure to untrustworthy websites in the 2016 U.S. election - PMC Exposure to untrustworthy websites in the 2016 U.S. election - PMC Exposure to untrustworthy websites in the 2016 U.S. election - PMC 3 AI deepfakes are poised to enter court proceedings at time of low trust in legal system AI deepfakes are poised to enter court proceedings at time of low trust in legal system AI deepfakes are poised to enter court proceedings at time of low trust in legal system 4 Authentication of Digital Audio Recordings Authentication of Digital Audio Recordings Authentication of Digital Audio Recordings 5 Understanding Digital Forensics: Process, Techniques, and Tools Understanding Digital Forensics: Process, Techniques, and Tools Understanding Digital Forensics: Process, Techniques, and Tools 6 Digital Forensics and the Chain of Custody: How Is Electronic Evidence Collected and Safeguarded? Digital Forensics and the Chain of Custody: How Is Electronic Evidence Collected and Safeguarded? Digital Forensics and the Chain of Custody: How Is Electronic Evidence Collected and Safeguarded? 7 People are trying to claim real videos are deepfakes. The courts are not amused People are trying to claim real videos are deepfakes. The courts are not amused People are trying to claim real videos are deepfakes. The courts are not amused 8 'Deepfake' audio evidence used in UK court to discredit Dubai dad 'Deepfake' audio evidence used in UK court to discredit Dubai dad 9 California Looks to Boost Deepfake Protections Before Elections California Looks to Boost Deepfake Protections Before Elections California Looks to Boost Deepfake Protections Before Elections 10 New laws to better protect victims from abuse of intimate images - GOV.UK New laws to better protect victims from abuse of intimate images - GOV.UK New laws to better protect victims from abuse of intimate images - GOV.UK

AI and the Law: Deepfakes, Eroding Trust, and the Legal Tightrope

Security Tends Toward Reactivity Securing organizations tends toward response if not just reactivity – backups are often at the end of the day and not usually tested to ensure that a restore would be valid; activity logs are aggregated and analyzed later on, so if anything happened, it’s discovered later; typical alerts are overabundant and take time to search through the bits and bytes in efforts to cut through the high amount of false positives to find any true culprit; and cameras come into play after something happens to see in hindsight what may have occurred. The search for real-time detection and analysis is evergreen. This real-time process has reached a highly respectable level with malware, email spoofing/phishing, and dangerous domain detection and protection, as demonstrated by the ubiquitous inclusion of features such as EDR, XDR, and in-depth email analysis in antimalware and internet security suites. A major driver of security solutions is the advancement in crime. Much of modern-day crime deals with stealing peoples’ data, so the tactics for that theft are improving. Additionally, the attack surface of data repositories is growing, making it easier for criminals to steal and harder for defenders to protect. Data is Treasure Data resembles treasure, and just like real treasure, it holds immense value but requires effort to be obtained and protected. · Importance: Just like some treasures are rare and irreplaceable, certain types of data are crucial and hold significant value. This could be financial data, intellectual property, personal information, or medical records. · Potential:  Like treasure chests, raw data might not reveal its true worth until analyzed and interpreted. Hidden patterns, trends, and insights can be extracted through data analysis, leading to valuable discoveries. · Vulnerability and Need for Protection: Similar to physical treasure, data requires robust security measures to prevent unauthorized access, breaches, and leaks. · Discovery and Exploration: The thrill of the hunt is a big part of the treasure hunting. Data exploration can lead to groundbreaking scientific discoveries, advancements in healthcare, and solving complex problems across various fields. Unlike a chest full of gold, data is constantly being generated and can be easily replicated. The true value lies in how it's managed, analyzed, and used. Just as a skilled treasure hunter needs tools and knowledge to find value, organizations need proper data management strategies and analytical tools to unlock the true potential of their data. Principles and Components of DDR According to a by ISC2 and Cloud Security Insiders: “Organizations are rapidly shifting workloads to the cloud, with 39% already operating more than half of their workloads in the cloud.” And in their study on adoption of cloud services, “Large companies saw an increase of 67%, and smaller companies increased cloud service use by 38%.” 2023 joint report according to Cloudwards To further underscore the value of data: “…data is the lifeblood of an organization…It’s not realistic to stop employees from collaborating with data or moving it to new places and systems to make better use of it. follows data across all these different assets.” Data Detection and Response Data Detection and Response (DDR) is an iteration of data security technology designed to address the challenges of protecting sensitive data. DDR focuses on the data itself, rather than just relying on perimeter defenses, and involves continuously monitoring and analyzing data flows in real-time across networks, endpoints, and cloud environments. The key principles of DDR include: · Classifying data based on its lineage and behavior, not just content alone · Tracking data movement across different assets and systems to maintain visibility · Taking real-time action to prevent data exfiltration and stop potential threats The primary components of DDR are: Monitoring: Continuously scan data activity across various sources (logs, endpoints, cloud environments). Detection: Identify anomalous data access and suspicious behavior using threat intelligence, analytics, and machine learning. Alerting: Generate timely and actionable alerts for security teams to investigate. Response: Take steps to isolate threats, contain breaches, remediate vulnerabilities, and recover compromised data. Data Detection – Your Resident Expert Data detection employs a combination of algorithms, statistical analysis, and machine learning to analyze vast amounts of data for irregularities. Security teams are : By establishing baseline patterns of normal behavior, data detection can swiftly identify deviations indicative of potential threats. This continuous monitoring of data streams, identifying those outliers, and flagging suspicious activities for further investigation is a complex dance necessary for proper discovery of relevant and real threats. Through this proactive approach, data detection acts as a frontline defense, enabling organizations to swiftly respond to emerging threats and safeguard their data assets. looking for anomalies “Anomaly detection, also known as outlier detection, identifies data objects or patterns that deviate from a dataset’s normal behavior. and identify anomalies that could signal potential threats.” Importance of Data Response Data response refers to the actions taken once potential threats or anomalies are detected in the data environment. This involves implementing real-time incident response measures to mitigate risks rapidly and effectively. DDR frameworks include incident response plans that empower organizations to respond swiftly to incidents, reducing the time it takes to detect and remediate threats and limiting the damage. Importance of proactive monitoring and continuous improvement. Proactive monitoring and continuous improvement are paramount in maintaining robust cybersecurity measures. By actively monitoring data systems, organizations can better detect and respond to threats before they escalate into full-scale breaches. Continuous improvement ensures that security measures progress alongside evolving threats and technological advancements. This proactive approach a) enhances resilience and b) fosters an organization-wide culture of vigilance and adaptability. Ultimately, it enables businesses to stay ahead of cyber threats and safeguard their data assets effectively. The Road Ahead Data is rapidly increasing in amount and its storage in cloud services. The better one’s ability to detect data motion and location, the better one can protect everyone involved. What’s your next step in protecting that treasure (whose whereabouts are known to you, not hidden)?

Data Security in the Cloud: Why You Need Data Detection and Response (DDR)

Channels in Go are a core part of the language's concurrency model, which is built around the principles of Communicating Sequential Processes (CSP). A channel in Go is a powerful tool for communication between goroutines, and Go's lightweight threads, enabling a safe and synchronized exchange of data without the need for traditional lock-based synchronization techniques. Channels serve 2 primary purposes in Go: : They ensure that data exchanges between goroutines are synchronized, preventing race conditions and ensuring data integrity. Synchronization : Channels facilitate the passing of data between goroutines, acting as a conduit for data flow in concurrent Go applications. Communication According to the Go specification: Channels are typed by the values they convey. This means that a channel has a specific type, and all communications through that channel must adhere to this type, ensuring type safety across concurrent operations. Key characteristics of channels: : Each channel is associated with a specific type of data it can transport. This type is determined at the time of channel creation. Typed : By default, sending and receiving operations on a channel block until the other side is ready. This blocking behavior is essential for synchronization, allowing goroutines to coordinate their execution flow. Blocking Operations : Go supports both buffered and unbuffered channels. Unbuffered channels do not store any values and are used for direct communication between goroutines. In contrast, buffered channels have a capacity and can hold a finite number of values for asynchronous communication. Buffered and Unbuffered Declaring channels In Go, channels are created using the function, which initializes and returns a reference to a channel. make The syntax for declaring a channel is: chanVar := make(chan Type) Where is the variable name of the channel, and specifies the type of data that the channel is intended to transport. It's important to note that the data type must be specified, as channels are strongly typed. chanVar Type Example of declaring a channel: messageChannel := make(chan string) This line of code creates a channel for transmitting strings, referred to by the variable . messageChannel Internals of Go channels Understanding the internals of Go channels involves delving into the struct, which forms the backbone of channel operations in Go. Let's break down how channels work under the hood, focusing on the struct and the initialization process when a channel is created with . hchan hchan make A channel in Go is more than just a conduit for communication between goroutines. It's a well-defined structure in memory, represented by . The struct is crucial for managing the state and operations of a channel, including sending, receiving, and synchronization. hchan hchan Here's an overview of its main elements: : This field holds the total number of elements currently in the queue. It's essential for understanding the channel's current load. qcount : Specifies the size of the circular queue. For an unbuffered channel, this is 0; for a buffered channel, it's the capacity defined at the creation time. dataqsiz : A pointer to an array where the channel's data elements are stored. The size of this array is determined by . buf dataqsiz : The size, in bytes, of each element in the channel. This ensures that the memory allocated for the channel's buffer is properly managed according to the type of data the channel is meant to hold. elemsize : A flag indicating whether the channel has been closed. Once a channel is closed, no more data can be sent on it, although data can still be received until the buffer is empty. closed : Points to a data structure that describes the type of elements the channel can hold. This is critical for maintaining type safety in Go's statically typed system. elemtype and : These indices manage the positions where the next send and receive operations will occur, respectively, enabling the circular queue functionality. sendx recvx and : Wait queues for goroutines that are blocked on receiving from or sending to the channel, respectively. These queues are implemented as linked lists. recvq sendq : A mutex lock to synchronize access to the channel, preventing race conditions when multiple goroutines interact with the channel concurrently. lock When a channel is created using the function, Go allocates memory for and initializes an instance of the struct. This process involves setting up the struct fields to their default values or the specified capacity for buffered channels. make hchan For example: ch := make(chan int, 10) This line creates a buffered channel of integers with a capacity of 10. We will delve deeper into the nuances of channel types and capacities in the upcoming sections of this article. Internally, Go does the following: Allocates a struct on the heap. hchan Sets to 10, reflecting the channel's capacity. dataqsiz Allocates an array of 10 integers ( will describe an integer type, and will be the size of an int on the current architecture) and assigns its address to buf. elemtype elemsize Initializes , , and to 0, indicating an empty channel. qcount sendx recvx Sets to 0, indicating that the channel is open. closed The initialization process ensures that the channel is ready for use, with a clear and safe protocol for sending and receiving data. This is crucial here. It's used to synchronize access to the channel, ensuring that concurrent operations are safe and that the channel's state remains consistent. lock Understanding the struct and the initialization process provides insight into how Go channels efficiently manage data exchange and synchronization between goroutines. This intricate design allows developers to leverage channels for robust concurrent programming without delving into the complexities of traditional thread synchronization mechanisms. hchan Operations on channel Once a channel is declared, it can be used for sending and receiving data. These operations are at the heart of channel based communication in Go. To send data to a channel, you use the channel variable followed by the send operator, and the value to be sent. The syntax looks like this: <- chanVar <- value In this example, a string is sent to the messages channel from within a new goroutine. The main function continues its execution without waiting for the send operation to complete, illustrating the non-blocking nature of send operations in buffered channels or when the receive is ready in unbuffered channels. Receiving data from a channel is done by placing the channel variable on the right side of the receive operator, . This operation is blocked by default. It waits until there's data to be received. The syntax for receiving data from a channel is: <- value := <-chanVar In this code snippet, the main function blocks at the receive operation until the goroutine sends a string to the channel. Once the message is received, it's stored in the variable and then printed to the console. messages msg Channel direction In Go, channels can be directional. This means a channel can be specified to only send or only receive values. Directional channels enhance type safety by ensuring a channel is used only for its intended purpose, whether it's to send data, receive data, or both. This concept is particularly useful in function signatures, where you want to enforce the role of a channel within the context of goroutine communication. A send-only channel can only be used to send data to a channel. Attempts to receive data from a send-only channel will result in a compilation error, ensuring that the channel's directionality is respected. The syntax for declaring a send-only channel is as follows: chanVar := make(chan<- Type) In this example, the function takes a send-only channel as an argument ( ) and sends a string into this channel. The main goroutine receives the string from and prints it. The function cannot receive data from , as it's a send-only channel, showcasing the enforcement of channel directionality. sendData sendCh chan<- string messageChannel sendData sendCh Conversely, a receive-only channel is used exclusively for receiving data. Sending data to a receive-only channel will result in a compilation error. The syntax for a receive-only channel is: chanVar := make(<-chan Type) Here, the function is designed to accept a receive-only channel ( ) from which it reads a message. The anonymous goroutine in the function sends a string to , which is then received by . This setup ensures that function cannot send data back through , adhering to its receive-only designation. receiveData receiveCh <-chan string main messageChannel receiveData receiveData receiveCh Capacity of a channel The capacity of a channel refers to the number of values that the channel can hold at a time. This is pertinent to buffered channels, which are declared with a specified capacity. The function is used to determine the capacity of a channel. cap() In this example, a buffered channel with a capacity to hold 5 integers is created. Using the function, we print out the capacity of , which is 5. cap() ch Length of a channel While the capacity of a channel is static, the length of a channel is dynamic and represents the number of elements currently queued in the channel. The function is used to find out how many items are currently stored in the channel. len() This code snippet demonstrates a buffered channel where two integers are sent into the channel. The function shows that the length of is 2, indicating two items are currently in the channel. len() ch Close operation on a channel Closing a channel indicates that no more values will be sent on it. This can be useful to signal to the receiving goroutines that the channel has finished sending data. The function is used for this purpose. close() After closing the channel , we attempt to read from it. The second value returned by the channel read operation indicates whether the channel is open or closed. In this example, after reading all the elements, becomes , signaling the channel is closed. ch open false Nil channel A nil channel is a channel with no reference. Both sending and receiving operations on a nil channel block forever, making nil channels useful for disabling a channel operation dynamically. Operations on a nil channel never proceed, making them distinct in their behavior compared to non-nil channels. For range loop on a channel The loop can be used to receive values from a channel until it is closed. This idiomatic way of reading from a channel ensures that all sent values are received. for range In this code, we iterate over a channel using a loop, printing each value received from the channel until it's closed. ch for range Advanced channel techniques Let's consider a scenario where multiple goroutines send data to a channel, and a single receiver processes the data in a synchronized manner. In this example, we spawn goroutines that send unique to the messages channel. The is used to wait for all sender goroutines to finish before closing the channel. The receiver goroutine uses a loop to receive messages until the channel is closed. numSenders messages sync.WaitGroup for range Next example showcases using the select statement to process data from multiple channels and a timeout feature. In the above code, we set up a pool of workers that receive from the jobs channel and send results to the channel. The function sends jobs to the workers and uses a statement to handle results with a mechanism. The timeout channel created with ensures that if results are not received within a certain period, the main program will not wait indefinitely. jobs results main select timeout time.After Both of these examples in this section present more complex use cases of channels in Go, demonstrating how channels can be used for synchronization, timeout handling, and concurrent processing patterns. Summary In wrapping up our exploration of Go channels, we've seen how these powerful tools facilitate communication and synchronization in concurrent programming. The behavior of channels varies depending on their state (open or closed) and type (buffered or unbuffered). To distill our understanding, let's review a summary table showcasing the outcomes of various operations across different channel states: Key Takeaways: : An attempt to send on a closed channel results in a panic, highlighting the need for careful channel management. Conversely, sending on a nil channel illustrates an indefinite block, a scenario typically indicative of a programming error or oversight. Send Operation : The nuanced behavior of receiving from a closed channel returning the default value if empty underscores Go's emphasis on safety and predictability in concurrent execution. Receive Operation : Successfully closing open channels ensures a clean state transition, while attempting to close a nil or already closed channel flags a clear misuse through a panic. Close Operation : These introspective operations provide visibility into the channel's current load and capacity, crucial for performance tuning and debugging. Length and Capacity Go channels offer a nuanced, powerful model for concurrency, balancing simplicity with depth. Understanding their behavior across different states and operations empowers developers to craft efficient, robust, and deadlock-free concurrent applications. As we've seen, mastery of channels is not just about knowing how to use them, but understanding their behavior under various conditions, ensuring our Go applications perform harmoniously even in the face of complex synchronization challenges.

Mastering Go Channels for Elegant Synchronization

As an AI expert, let me be blunt - cyber threats are a massive, unrelenting risk that no business with an online presence can afford to ignore these days. Whether you're running an e-commerce site, logistics operation, or tech firm, phishing, ransomware attacks, or data breaches - these scourges can cripple a company, leading to devastating financial losses, irreparable brand damage, and even regulatory penalties that could bury you. I've personally seen the aftermath of cyber strikes on SMBs with limited security resources and tight budgets. In those cases, traditional cybersecurity just doesn't cut it against today's attack of threats. But that's exactly where artificial intelligence (AI) emerges as a total game-changer. discovered that organizations leveraging AI and automation extensively saved a mind-blowing $1.76 million on average compared to their less AI-savvy counterparts. IBM's 2023 Cost of Data Breach Report And I can testify that AI-powered cyber defense solutions truly earn those savings. By unleashing cutting-edge AI techniques like machine learning, deep learning models, and natural language processing, modern AI cybersecurity acts like an indispensable digital force multiplier reinforcing our defenses against even the wiliest adversaries. As an AI expert, I've deployed these capabilities for clients - it's like having an entire squad of cybersecurity sentries patrolling the digital premises 24/7 to keep the bad actors at bay. Moving forward, I don't believe adopting advanced AI will be a luxury for organizations - it's a flat-out necessity we can't ignore. Whether you're a small business or a giant enterprise, you'd better get proactive and start leveraging AI's offensive cybersecurity potential to safeguard your vital digital assets and ensure continuous operations. In my opinion, AI is quickly separating the cybersecurity haves from the have-nots. AI's Cyber Offense: How It Turbocharges Threat Detection I can't overstate how AI is revolutionizing cybersecurity by unleashing cutting-edge threat detection and response capabilities never before possible. Just look at the rapidly growing market - projects that AI in the cybersecurity sector will explode from $22.4 billion in 2023 to $60.6 billion by 2028. And there are good reasons behind that meteoric growth. MarketsandMarkets Machine learning algorithms excel at quickly identifying patterns and anomalies within massive datasets - making them invaluable assets for recognizing potential cyber threats amid tons of network traffic, user activity logs, and other data. By training these algorithms on historical data, we can have them recognize known attack signatures and deviations signaling new threats before they can wreak havoc. One study I find particularly compelling is from MDPI, showing machine learning models achieved up to 99% malware detection accuracy using techniques like Decision Trees, Convolutional Neural Networks, and Support Vector Machines. That's the level of precision AI can deliver against emerging threats. There are deep learning models that I've deployed to rapidly analyze and classify malware samples with incredible accuracy. The latest deep-learning architectures can see patterns invisible to the naked eye. However, some of the most exciting AI cybersecurity applications my team has tested involve reinforcement learning for adaptive threat response. These self-learning systems can dynamically adjust defensive actions on the fly based on their experiences against evolving attack vectors. Allied Market Research predicts reinforcement learning will see explosive 41.5% annual growth from 2023 to 2032, with cybersecurity as a prime use case. Let's not overlook natural language processing (NLP) for automating incident analysis. I've personally overseen NLP deployments that can rapidly comprehend human language data from reports, intelligence feeds, and more - streamlining our incident response workflows and prioritizing genuine threats over false positives. Combining all these AI disciplines is delivering incredible offensive capabilities to counter even the trickiest cyber adversaries. As an expert and founder in this space, I truly believe AI's ability to autonomously detect and shut down threats before impact will utterly transform cybersecurity over the coming years. Why UEBA and User Behavior Monitoring Matter (And How AI Excels) The cyber risk from malicious insiders or compromised user accounts is one we can't overlook either. That's why I consider user and entity behavior analytics (UEBA) powered by advanced AI/ML models to be paramount for comprehensive, 360-degree security visibility. At a basic level, these AI-driven UEBA solutions leverage machine learning to establish baselines modeling each user's normal behavior based on their activity patterns. Once those baselines are established, the AI can continuously monitor for deviations that may indicate insider threats, privilege abuse, compromised accounts, and other insider risk vectors. What makes this application of AI so powerful is its ability to dynamically score the risk level of unusual user actions based on contextual factors like their role, access privileges, the sensitivity of data/systems involved, and more. This enables real-time flagging of potentially malicious insider activity for further investigation - something that was highly prone to human error with traditional, rigid rule-based approaches. Another area where I've seen UEBA with AI shine is detecting compromised privileged user accounts, which are prime targets for external threat actors seeking a foothold. By understanding the unique behavior patterns of different privileged user roles, AI can rapidly identify the subtle anomalies of a privileged account under hostile control - giving our teams a chance to shut down that breach before catastrophic damage occurs. In my opinion, deploying AI for insider and privileged user threat detection by understanding human behavior is an absolute game-changer. Monitoring our people and accounts is just as vital as external threat monitoring - without it, we're leaving a gaping hole in our defenses that skilled adversaries will inevitably find and exploit. Of course, my teams always emphasize that AI-powered UEBA is just one layer in a comprehensive, defense-in-depth strategy. But it's a crucial layer that lets us get in front of insider incidents before they escalate - buying time that can mean the difference between a contained issue and a full-blown crisis. The Flip Side: Defending Against AI-Powered Threats As amazing as AI's cybersecurity offensive capabilities are, I have to address the flip side too: hackers are actively working on ways to weaponize AI against us. Adversarial attacks designed to bypass, poison, or steal our AI models are a clear and present danger we can't ignore. One disturbing forecast from research analysts indicates: Up to 30% of AI cyberattacks will leverage training data poisoning, model stealing, or adversarial examples to attack AI-powered systems. But we're not defenseless - there are proven ways to harden AI systems against these tactics if we invest in robust AI security controls from the start. Approaches like adversarial training, where models are intentionally exposed to deceptive inputs during the learning process, can improve resilience against evasion. Differential privacy techniques can also obscure model behavior to thwart extraction attacks. We follow best practices around secure AI development Life Cycle, subjecting our models to rigorous adversarial testing and employing defensive AI techniques to stay ahead of emerging threats. I don't mean to sound alarmist, but the cybersecurity industry at large needs to prioritize battling AI-powered threats just as much as leveraging AI for defense. It's a two-way street - we must secure the AI models underpinning our cyber defenses with the same vigor we apply AI offensively. Neglecting AI security WILL be exploited by sophisticated hackers. So while I'm incredibly confident about AI's potential to revolutionize cyber threat detection and response, I'm also pragmatic. We simply can't ignore the dark side and must proactively invest in developing and deploying AI securely to maintain a decisive strategic advantage. Integrating AI Across Security Operations Of course, AI alone is not a silver bullet for cybersecurity. Its true power comes from thoughtful integration across existing processes, teams, and technology stacks. As an enthusiastic AI Expert, I emphasize this force-multiplier effect to clients constantly. AI should augment and empower human expertise, not replace it entirely. That's why the most successful deployments I've seen use AI to offload tedious, repetitive tasks to machines - freeing up skilled personnel to focus on higher-level analysis, investigation, and decision-making that requires human judgment. That's also why AI is rapidly becoming essential for SOC analysts and incident response teams: by automating alert triage, log analysis and basic response workflows, AI drastically reduces burnout from low-value "grunt work" while improving overall efficiency. Research from Capgemini revealed organizations using AI for security operations saw a 12% increase in breach detection rates and an impressive 19% reduction in time-to-respond. Those are major gains I've validated through my own startup's AI deployments. AI and machine learning are also driving huge advances in security orchestration, automation, and response (SOAR) tools that streamline entire incident response processes. With SOAR, AI handles tasks like correlating disparate security signals, surfacing relevant threat intelligence, and even executing semi-automated response playbooks. It's a huge step towards smarter, more autonomous incident handling. Ultimately though, AI is a complement to skilled human teams, not a wholesale replacement. Cybersecurity operations will always require human expertise, analysis, and judgment. But by intelligently augmenting our people with AI automation and assistance, we can concentrate their talents on higher-impact duties while optimizing productivity and efficiency. I already see this human-machine symbiosis enhancing the industry's ability to outpace threats. As AI integration continues maturing, I'm confident embracing this collaborative model is the only way we'll realistically close the cybersecurity skills gap plaguing our field. AI's Cybersecurity Superpowers By rapidly analyzing network data, user activities, system logs, and more, AI can identify malicious behavior patterns and potential threats with superhuman speed - giving us an early warning to shut down attacks before escalation. Threat Detection & Prevention: While phishing lures and malware grow more devious, AI algorithms can precisely distinguish malicious files and emails from legitimate ones by evaluating content, sender metadata, behavioral signals, and other artifacts that often evade human analysis. Phishing and Malware Detection: Rather than rigid rules, AI lets us dynamically model normal user activity baselines. It can then rapidly detect anomalies indicative of insider threats, compromised accounts, data exfiltration attempts, and more - all in real-time. User Behavior Analytics: I've deployed AI to prioritize vulnerabilities based on their risk and exploit potential, automating the triaging process. AI can also correlate vulnerabilities with active threats, showing us which exposures need patching first. Automated Vulnerability Management: Modern AI-powered SIEM solutions can correlate security events across networks, endpoints, and other sources to rapidly identify genuine incidents amidst the noise. AI also streamlines response workflows based on alert contexts. Smart SIEM & Incident Response: Across incident triage, alert analysis, threat hunting, and more, AI is automating repetitive, labor-intensive security tasks - amplifying the impact of human personnel while reducing burnout and fatigue. Security Ops Automation: I could go on, but you get the idea. As an AI expert and entrepreneur, I've seen these capabilities utterly transform cybersecurity operations into an amazingly proactive, autonomous counter-threat force. And AI's offensive cyber capabilities are still rapidly evolving. The Way Forward: Practical AI Cyber Defense But when I gaze into the future of AI-powered cybersecurity, I see incredible potential beyond just better threat detection and response. Used wisely, I believe AI can fundamentally reshape and elevate our entire profession into a truly proactive, predictive security posture - finally outpacing even the wiliest cyber adversaries. However, realizing this AI-driven future won't happen automatically. It requires us as an industry to thoughtfully and securely integrate AI across our people, processes, and technology stacks with a keen focus on augmenting human expertise and capabilities rather than wholesale automation. At an operational level, this will mean embracing increasingly autonomous security workflows offloaded to AI systems - handling tedious triaging and basic response while escalating only the most critical issues to skilled analysts for analysis and key decisions. It's a pragmatic balance of human and machine capabilities. We'll also need a new class of cybersecurity professionals fluent in interpreting AI system outputs and translating them into strategic guidance. These AI-savvy analysts must be adept at not just using but explaining and justifying AI system behaviors and decisions to leadership stakeholders. It plays a crucial role in building organizational trust and accountability around AI systems. Additionally, organizations must adopt secure AI system development lifecycles, proactively battling adversarial AI threats through approaches like robust model training, adversarial testing, privacy-preserving analytics, and defensive AI architectures. We simply can't leave our AI systems vulnerable to adversarial subversion. Thankfully, I see positive momentum towards these goals across the AI security startup ecosystem and cybersecurity community. But we must keep pressing forward holistically - investing in skilling our people, refining our processes, and pioneering new AI-centric technologies securely and ethically. Only then can we responsibly wield AI's incredible cyber defense potential without enabling worse nightmares. Despite all the work ahead, I'm hugely optimistic as an AI expert and founder. I've already deployed these transformative AI capabilities in the field, experiencing their game-changing impact firsthand. AI gives us a truly decisive advantage in combating even the most sophisticated threat actors - if we successfully fuse machine intelligence and human expertise. It won't be easy, but I'm convinced the payoff will be a safer, more secure digital world actively hardening itself against cyber threats in real-time. A world where novel attacks are neutralized before impact, where vulnerabilities are patched before exploitation, and where malicious insiders and compromised accounts are stopped in their tracks. Best of all, AI cyber defense done right should free up our human personnel for higher-impact strategic work, not tedious data grunt work. AI should empower us to be the cybersecurity experts, creative problem-solvers, and strategic leaders we always aspired to be - not menial attack analysts. So to my fellow AI professionals, founders, and innovators - the path ahead is clear. Let's blaze this trail TOGETHER, pioneering the new era of practical, intelligent, and ethical AI-powered cyber defense. The future of digital safety depends on us doing this right. And with our unique combination of human ingenuity and machine intelligence, I'm confident we're up to the immense challenge. Let's get to work.

Applications of Artificial Intelligence in Cybersecurity: Boosting Threat Defense System

AI has radically reshaped predictive analytics, making forecasting more precise than ever before. Businesses in all industries are taking advantage, capitalizing on AI’s wide-ranging capabilities to anticipate future trends and prepare for what’s coming around the corner. The accuracy of any forecast hinges on a number of factors: the sophistication of , the robustness of predictive models, and the quality and quantity of the data they’re trained on, to name a few. AI algorithms is a generative pre-trained model specifically designed for predicting time-series data. It’s a powerful asset that will allow you to forecast future outcomes with greater precision, giving you the foresight to capitalize on upcoming opportunities and dodge possible pitfalls. Nixtla’s TimeGPT Below, we’ll put TimeGPT under the microscope, analyze it from all angles, and show how you can integrate it with your database to maximize the full potential of predictive analytics. What is TimeGPT? TimeGPT is a pre-trained generative model for time-series data forecasting. Its core functionality is to predict future values in a time series based on historical data. The application processes data sequentially, using a windowing technique to analyze historical data points and recognize patterns. This enables it to predict future values by extrapolating these identified patterns. An overview of TimeGPT : Nixtla's TimeGPT applies pre-trained models to time-series data, using the data collected to predict new datasets without specific training, a departure from traditional models. Innovative Pre-training : TimeGPT excels in identifying complex patterns in time series data, surpassing many traditional models in trend, seasonality, and anomaly detection. Advanced Pattern Recognition : By using advanced neural networks, TimeGPT offers efficient forecasting, reducing the time and resources typically needed for time series analysis by eliminating dataset-specific retraining. Efficient Forecasting : Unlike static traditional models, TimeGPT has the potential for ongoing learning and adaptation, maintaining accuracy as new data emerges. Continuous Learning and Adaptation TimeGPT’s features and benefits As one of the leading predictive analytics models, TimeGPT offers advanced capabilities that provide a variety of unique features, making it a powerful asset for managing complex forecasting scenarios. These include: Anomaly Detection Leveraging predictive modeling techniques, TimeGPT excels in identifying unusual patterns or outliers in time series data, which are often indicative of significant but unexpected events. Feature: This feature is crucial for businesses and analysts in quickly detecting and responding to unusual occurrences, ensuring that forecasting models remain accurate and reliable, even in the presence of aberrations. Benefit: Exogenous Variables Handling TimeGPT can incorporate external factors or exogenous variables (like weather conditions for retail sales forecasting) into its analysis. Feature: This ability allows for more nuanced and accurate forecasts by considering the impact of external elements, which is particularly useful in scenarios where external factors significantly influence the data trends. Benefit: Fine-tuning While TimeGPT is a pre-trained model, it offers the capability for fine-tuning to adapt to specific datasets or unique forecasting needs. Feature: This feature enables users to tailor the model more closely to their specific requirements, enhancing the accuracy and relevance of the forecasts in specialized contexts. Benefit: Historical Forecasting TimeGPT can analyze past data to generate historical forecasts, allowing users to understand how the model would have predicted past events. Feature: This is particularly beneficial for validating the model’s accuracy and reliability, giving users confidence in its future predictions. Benefit: Managing complex forecasting scenarios TimeGPT’s features are especially beneficial when handling complex scenarios such as holidays, special dates, and irregular timestamps. These events can significantly impact trends in time series data. TimeGPT’s ability to detect anomalies and incorporate exogenous variables helps in accurately forecasting around these dates, considering their unique impact. Holidays and Special Dates: Time series data can sometimes be irregular or have missing points. TimeGPT’s sophisticated algorithms can handle such irregularities, ensuring consistent and accurate forecasting even with non-uniform data. Irregular Timestamps: Deploying TimeGPT models in MindsDB MindsDB is the platform for building custom AI from enterprise data. It works by connecting any source of data with any AI/ML model or framework and automating how real-time data flows between them. MindsDB bridges data sources with advanced AI/ML models like TimeGPT. This allows you to easily create, train, and manage these models, making it particularly beneficial to those who have minimal experience in machine learning. Integrating TimeGPT into MindsDB begins with creating a TimeGPT ML engine using the below statement and providing the TimeGPT API key. CREATE ML_ENGINE timegpt_engine
FROM timegpt
USING timegpt_api_key = ‘timegpt-api-key’; Next, users can employ the statement to create, train, and deploy the TimeGPT model. CREATE MODEL CREATE MODEL model_table
FROM data_source
  (SELECT * FROM data_table)
PREDICT column_to_be_predicted
ORDER BY date_column
GROUP BY column_name, column_name, ...
HORIZON 3 -- model forecasts the next 3 rows
USING ENGINE = 'timegpt_engine'; Now you can query the model to get forecasts: SELECT m.date_column, m.column_to_be_predicted
FROM data_table AS d
JOIN model_table AS m
WHERE d.date_column > LATEST; With MindsDB, you can automate forecasts and generate them constantly for any downstream usage. It is especially useful when you have dynamic data that gets updated regularly. Then, the model should be fine-tuned with the latest data to provide as accurate forecasts as possible. Here, we create a job that fine-tunes the model with the latest data and saves daily forecasts in a database table. CREATE JOB automate_forecasts (

   FINETUNE model_table
   FROM data_source
      (SELECT * FROM data_table);

   CREATE TABLE daily_forecasts_{{START_DATETIME}} (
      SELECT m.date_column, m.column_to_be_predicted
      FROM data_table AS d
      JOIN model_table AS m
      WHERE d.date_column > LATEST
   )
)
EVERY 1 day; To see a complete example, visit our . You can find more . docs page here forecasting use cases here How to use TimeGPT in predictive analytics Companies must be agile enough to react to minute-by-minute market changes. Using predictive analytics techniques, such as those enabled by TimeGPT, will help keep you a step ahead and even give you a competitive advantage. Businesses can utilize TimeGPT in several key areas, including: Financial Forecasting TimeGPT’s proficiency in processing and interpreting time-based data leads to more precise forecasts in the fluctuating realm of finance. Its ability to identify current trends and subtle nuances within financial datasets helps to create more informed investment strategies. Demand Prediction In industries like retail and manufacturing, TimeGPT aids in predicting product demand. This is crucial for inventory management, supply chain optimization, and strategic planning. By analyzing historical sales data and market trends, TimeGPT can forecast future demand patterns, helping businesses to prepare and adapt effectively. Anomaly Detection The model's approach to sequential learning proves advantageous in detecting anomalies, like spotting unusual patterns in network traffic for cybersecurity or identifying machine performance irregularities for predictive maintenance. This function is crucial in averting potential problems before they intensify. Predictive analytics use cases Here are just a few ways to use TimeGPT for accurate predictive analytics. TimeGPT performs well compared to traditional methods in stock price prediction, aiding investors and analysts. Stock Market Analysis: : It predicts healthcare trends, which is important for resource allocation and planning. Healthcare Trend Anticipation TimeGPT predicts future demands in logistics, improving efficiency and reducing costs. Supply Chain Optimization: It forecasts retail buying trends, aiding in inventory management and targeted marketing. Retail Customer Behavior Analysis: TimeGPT identifies potential equipment failures, minimizing downtime and maintenance costs. Manufacturing Predictive Maintenance: TimeGPT in the real world Global business is more complex and interconnected than ever. One false move in a single industry can have a ripple effect in several others, making it crucial for companies to leverage predictive analytics to foresee, prepare, and react to future events. With its advanced time-series analysis, deep learning capabilities, adaptive learning, and customizable scalable models, TimeGPT has proven to be an effective tool for more accurate predictive analysis. Integrating TimeGPT with MindsDB creates a powerful combination that enhances the precision and applicability of predictive insights directly within business databases. This gives you the agility to anticipate and react swiftly to a rapidly evolving global market. How would TimeGPT perform on your company’s data? Check out the . tutorial here Also published . here

How TimeGPT Transforms Predictive Analytics with AI

টাইমজিপিটি কীভাবে AI এর সাথে ভবিষ্যদ্বাণীমূলক বিশ্লেষণকে রূপান্তরিত করে

Wie TimeGPT Predictive Analytics mit KI transformiert

Cómo TimeGPT transforma el análisis predictivo con IA

Comment TimeGPT transforme l'analyse prédictive avec l'IA

TimeGPT कैसे AI के साथ पूर्वानुमानात्मक विश्लेषण को बदलता है

TimeGPT が AI を使用して予測分析をどのように変革するか

TimeGPT가 AI를 통해 예측 분석을 혁신하는 방법

Como o TimeGPT transforma a análise preditiva com IA

Как TimeGPT преобразует прогнозную аналитику с помощью ИИ

TimeGPT, Yapay Zeka ile Tahmine Dayalı Analitiği Nasıl Dönüştürüyor?

Cách TimeGPT biến đổi phân tích dự đoán bằng AI

TimeGPT 如何利用 AI 改变预测分析

**AUSTRALIA, South Australia, March 8th, 2024/Chainwire/--**Bavarsis: Pioneering Innovation in the Cryptocurrency Market. March 2024, Australia – In the dynamic landscape of investment, Bavarsis emerges as a beacon of innovation and opportunity, offering a realm of innovation and potential that transcends mere user-friendliness. Leveraging the capabilities of Artificial Intelligence (AI) and trading bots, Bavarsis facilitates rapid arbitrage trading, presenting a versatile framework, diverse cryptocurrency options, transparency, and a steadfast dedication to enduring viability. Delving into the core of Bavarsis unveils the myriad reasons why it stands as a premier option for contemporary investors. Bavarsis, a leading player in the cryptocurrency market, proudly announces remarkable achievements in its recent financial report. The company boasts a substantial increase in its crypto-assets by 4.1% over the past month, indicative of robust growth and effective investment strategies.Furthermore, Bavarsis reports an impressive 24% growth on its assets, attributed to the dynamic nature of the cryptocurrency market. These results underscore the company's sound investment decisions and its adeptness at capitalizing on market trends effectively. "We are delighted to share these positive outcomes with our partners and clients," stated the Bavarsis team. "Our commitment to strategic investment and prudent decision-making has not only strengthened our position in the market but has also yielded significant benefits for our valued clients." In a testament to its commitment to expansion and global presence, Bavarsis inaugurated several new offices in key regions including Russia, Taiwan, China, and Europe. This strategic move underscores the company's intent to establish a strong foothold in emerging markets and enhance accessibility to clients worldwide. Moreover, Bavarsis marks a significant milestone with the launch of its new exchange platform, aptly named Bavarsis Exchange. The launch of this exchange reflects the company's dedication to innovation and providing cutting-edge solutions to its clientele. "We are thrilled to announce the launch of Bavarsis Exchange, a platform designed to meet the evolving needs of cryptocurrency traders and investors," remarked the Bavarsis team. "With advanced features and a user-friendly interface, Bavarsis Exchange aims to offer a seamless trading experience to our users." As Bavarsis continues to expand its presence and explore new opportunities in the cryptocurrency space, the company remains steadfast in its commitment to fostering long-term partnerships and delivering value to its stakeholders. In addition to its impressive achievements and expansion initiatives, Bavarsis operates under the jurisdiction of two key regions, further solidifying its strategic positioning and growth prospects. The company is registered in two jurisdictions, namely Hong Kong and Australia, each offering distinct advantages for its operations. Hong Kong, renowned for its business-friendly environment and regulatory framework, provides an ideal setting for setting up and scaling an exchange platform like Bavarsis Exchange. Leveraging the robust infrastructure and regulatory support in Hong Kong, Bavarsis aims to expand its exchange operations and cater to a global clientele seamlessly. On the other hand, Australia emerges as an ideal jurisdiction for Bavarsis' arbitrage activities, primarily due to its favorable tax policies. With low taxes on arbitrage activities, Australia presents lucrative opportunities for Bavarsis to maximize its profitability and optimize its investment strategies effectively. By capitalizing on the advantageous tax environment in Australia, Bavarsis can enhance its competitiveness and solidify its position as a leader in the cryptocurrency market. For more information about Bavarsis and its latest developments, visit their official website at bavarsis.com Contact Bavarsis exchange conquers new heights Bill Bavarsis Office@bavarsis.com +61 451 602 190 This story was distributed as a release by Chainwire under HackerNoon’s Business Blogging Program. Learn more about the program here.

BAVARSIS EXCHANGE Conquers New Heights

How do you choose the most appropriate model for your application? An analysis of evaluating and building applications on open-source large language models. The computational complexity of AI models is growing exponentially, while the compute capability provided by hardware is growing linearly. Therefore, there is a growing gap between those two numbers, which can be seen as a supply and demand problem. On the demand side, we have everyone wanting to train or deploy an AI model. On the supply side, we have Nvidia and a number of competitors. Currently, the supply side is seeing earnings skyrocket, and the demand side is stockpiling and vying for access to compute. It’s a simple formulation but one that works well to explain the current situation in AI. If anything, the generative AI explosion in the last couple of years has validated this formulation. How can this supply vs. demand imbalance be resolved then? One way would be to introduce , and some people are working on that. If you are not one of them, the other option would be to work on the software side of things. That is, to try and come up with ways to make AI model training and inference less computationally intensive. more AI hardware with more capabilities This is what CEO and co-founder Yonatan Geifman has been working on since 2019. Deci is focused on building AI models that are more efficient and accurate. Initially, , later adding a range of language models to its portfolio. Deci Deci released computer vision models Interestingly, most of these models are open source. We caught up with Geifman to explore the landscape of open-source AI, aiming to address some key questions. Do the benefits of open source and the criteria to consider in the classical build vs. buy dilemma apply to AI large language models (LLMs)? And how can LLMs be evaluated and tailored to the specific needs organizations have for their applications? https://youtu.be/1GD9J84Bw34?embedable=true Open source Large Language Models are different To begin with, let’s examine what “open source” means in the context of LLMs. Open-source software has been around for a long time and is sufficiently well understood. , and different licensing schemes have their peculiarities. Open source has gone through some turbulence Still, in software, things are relatively straightforward. It’s all about the source code and its availability. Having the source code and the right dependencies and environment means being able to build and deploy software. But what about AI models? Yes, we also have the source code. However, there are additional artifacts on the way from source code to deployment. We have the dataset used to train the model. We have the training process and . Some model releases also include metadata such as metrics and . Can the same open-source definition apply? weights for the source code model cards Geifman believes that the most important artifacts are the source code and the weights, as having those shared enables people to use and build on top of other people’s work and collaborate around those derivatives that are open source. Metadata is nice to have but not essential in his view. But there’s also the crucial issue of data and training process. Deci, like Mistral, is not sharing data or details about the process of training their open-source models. Geifman noted that there are trade secrets around those. To get a model that works well, it’s not just about having the right data but also about using the right parts of the data in the right way, he added. Geifman thinks that sharing the data and the training process is significantly less important than sharing the model itself. By contrast, that “trained on the same dataset for long enough, pretty much every model with enough weights and training time converges to the same point.” OpenAI’s James Betker claims What probably everyone will agree to is the fact that open-source software and open-source models are not exactly the same thing. As well as the fact that if open sourcing of model architectures and software had not happened, AI would not be at the point where it is today. Either way, there is a whole spectrum from having everything open to having everything closed, and most models fall somewhere in between. Geifman noted that there is a tendency in the community to use the term models instead of open-source models. open weights Open source Large Language Models vs. GPT-4 and Claude Open weights or open source, however, what’s probably more important is how these models can be used by organizations. This is where things look more like traditional software. Organizations have many choices. They can use open-source models hosted in their own infrastructure and in some third-party providers' infrastructure, or they can use hosted, closed-source models such as GPT-4 or Claude 3 via an API. What Geifman sees is that many organizations embark on their AI journey by leveraging closed-source model APIs. He thinks the reason is it’s easier to get started this way, as there is less investment in time and expertise. However, over time, that may not be the more sensible approach. Prices may fluctuate, but as closed-source APIs come with a premium, costs will start adding up. But this may not even be the most serious drawback. Closed source models are black boxes controlled by a third party, and there are implications to this. When things go well, using a thing party model may mean lowering the bar to entry and having peace of mind. But the flip side is that if the model is compromised, goes down, or suffers from a bug or degradation in performance, applications that use it will suffer the consequences. Similar effects will also occur even when updating models, as their behavior may change inconsistently. Prompt engineering is brittle enough as it is, so adding this parameter makes building applications on top of closed-source LLMs an exercise in non-determinism. There are also privacy and security concerns to be aware of, as sensitive data may be exposed. This is the reason why some organizations are not only refraining from using closed-source models but outright banning their employees from using them, too. Large Language Model customization: Fine-tuning vs. RAG Most of the above is not specific to AI models or LLMs. What is specific, and more often than not required for real-world applications, is customization. Whether it’s via , LLMs need to be tailored to users' needs and datasets. fine-tuning or RAG Even though fine-tuning and RAG aim to achieve similar goals, the way they work is quite different, as Geifman pointed out. Fine-tuning changes the actual model for a specific data set, while RAG focuses on injecting data into the generation phase. Geifman finds that with RAG, there is not a lot of difference if you’re looking to apply it to an open-source or a closed-source model. Fine-tuning, on the other hand, will be much easier to do for an open-source model. In the debate of fine-tuning versus RAG, Geifman thinks the future is a combination of both. “There are some benefits in using RAG, mostly on data privacy. You don’t want to train the model on private data, but in RAG you can supplement the model with private data in the prompt. That’s one of the benefits of using RAG, and also the fact that you don’t need to fine tune the models all the time in order to have updated data”, Geifman said. It’s worth mentioning that both fine-tuning and RAG are constantly evolving, with techniques such as and , respectively. Overall, however, Geifman thinks that building more complex use cases will be much easier to do in open source versus closed source in the future, especially when the accuracy gap is bridged. LoRA+ Corrective RAG The reference to accuracy brings us to another multi-variant equation, namely, how to evaluate LLM performance. At the moment, as Geifman mentioned, it seems that closed-source models such as OpenAI’s GPT-4 or Anthropic’s recently released Claude 3 have an edge in terms of accuracy over open-source models. But there is a lot of nuance here. Evaluating Large Language Models Geifman predicted that the gap will be closed by the end of 2024. However, as has recently been demonstrated, carefully fine-tuned models specializing in specific tasks can outperform larger models for these tasks. LoRA Land, a collection of 25 fine-tuned open-source Mistral-7B models, was shown to consistently outperform base models by 70% and GPT-4 by 4-15%, depending on the task. What’s more important, however, is evaluating performance holistically. Typically, people tend to equate LLM performance with accuracy. While accuracy is an important metric, there are many other parameters that need to be considered as well. These all revolve around the nature of the application for which the LLM will be used. Even a typical use case such as document summarization, for example, has different requirements depending on the type of document and what the summary is used for. Latency and throughput are also very important, again depending on the application requirements. Some applications may be more tolerant than others. And then there’s , obviously. Even though this is not a metric per se, it’s something that needs to be factored in as well. cost Geifman’s recommendations? First, if you need speed, you’d better work with smaller, fine-tuned, tailored models for your applications. Second, for most B2B applications, you don’t need GPT-4-level accuracy. If you’re trying to summarize documents, you can do it with . mid-sized language models with good performance, lower latency, and better value for money As to how to evaluate models holistically, that’s still an unsolved problem. The two main alternatives at the moment are human evaluation and using an LLM to evaluate another LLM. Both have pros and cons and as usual, there are lots of nuances and blind spots. You might be tempted to note that it’s all fine and well, but not everyone needs to be familiar with all the nuances of LLM evaluation. There is always the , which has become the de facto evaluation ground for all LLMs. Just like any other benchmark, however, these metrics are only indicative. Hugging Face leaderboard The bottom line is that if you want to assess whether a model is cut out for the task at hand, you can stand on the shoulders of giants, but you’ll have to put in the work yourself. And that means formulating your requirements, developing realistic evaluation scenarios, and putting LLMs to the test. Market outlook In the end, the fact that best practices and principles from other areas also apply at least partially to LLMs is probably good news for organizations looking to incorporate LLMs in their applications. As Geifman noted, the excitement from LLMs and their capabilities and the potential of integrating them in enterprises across many verticals has energized the market. Geifman shared that Deci’s AI platform has seen its market more than triple in 2023. Deci’s market is divided into two segments, each with its own characteristics. One of them is computer vision, which is more mature. People know what they need, and they are looking towards production use cases with multiple use cases per company and team, as Geifman put it. Then, there is the GenAI segment, in which people are more in an exploratory phase. Geifman believes 2024 will be the year we’ll see many GenAI-based projects going into production. He sees document summarization and customer care as the dominant use cases. https://pod.co/orchestrate-all-the-things-podcast-connecting-the-dots-with-george-anadiotis/evaluating-and-building-applications-on-open-source-large-language-models-featuring-deci-ceo-co-founder-yonatan-geifman?embedable=true Deci addresses both computer vision and GenAI with open-source models. includes SuperGradients, a library for training PyTorch-based computer vision models, DeciLM 6B and DeciLM 7B, which Geifman noted offers similar accuracy to Mistral 7B with better efficiency and performance, the DeciCoder 1B and DeciCoder 6B code generation models as well as the DeciDiffusion text-to-image model. Deci’s offering These models are free to use, with additional tools to customize, fine-tune, and deploy them. Deci’s deployment environment for LLMs is called Infery-LLM. It’s an open-core approach in which the models are open source, and the value-add tools are commercial. To showcase the value that Infery-LLM brings, Geifman shared that running DeciLM 7B with and without Infery-LLM gets 4,5x vs. 2x the performance of Mistral 7B, respectively. Deci is also working on releasing more LLMs with more parameters and better performance. Also published . here

INDICATIVE

indicative.com

-- employees

Since n.d.

COMPANY RANKING

INDICATIVE

EVERGREEN INDEX #6037

Representing this company? Click here to claim and customize this page!

Indicative WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!

INDICATIVE

indicative.com

-- employees

Since n.d.

COMPANY RANKING

INDICATIVE

EVERGREEN INDEX #6037

RELATED COMPANIESRANK

STORIES ABOUT INDICATIVE

Representing this company? Click here to claim and customize this page!

Indicative WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!