Subscribe to our daily newsletter to keep in touch with the subjects shaping our future.

Experts Alarmed by People Uploading Their Medical Scans to Elon Musk's Grok AI

Sam Altman Points Out That Elon Musk's "Anti-Woke" Grok AI Seems to Actually Support Kamala Harris

Elon Musk Recommends Feeding Your Medical Scans Into His Grok AI

Meta&#x27;s AI chief Yann LeCun suggested Elon Musk leverage a new FAIR paper on Contextual Positional Encoding to enhance xAI&#x27;s Grok, despite their ongoing public feud.

Meta AI Chief Yann LeCun Says Elon Musk Could Use This New Paper To Improve xAI's Grok Amid Ongoing Feud With Tesla CEO - Meta Platforms (NASDAQ:META) - Benzinga

X platform&#x27;s Musk has announced big update to his Grok AI chatbot coming soon

Elon Musk Teases Major Grok AI Upgrade, Here's What It Will Do

It was founded in 2016, when it trademarked the name Groq. Last November, when Elon Musk’s own AI model, also called Grok — but spelled with a “k” — was gaining traction, the developers behind the original Groq published a blog post calling out Musk for the choice of name:

Groq AI model goes viral and rivals ChatGPT, challenges Elon Musk’s Grok

Grok

grok

Elon Musk, the entrepreneur and CEO of X, recently unveiled Grok, an AI chatbot designed to provide real-time answers to a wide range of questions, all while maintaining a personality and a sense of humor. This innovative concept has garnered considerable attention, with Musk touting Grok's unique ability to tap into the real-time stream of commentary and analysis on the social network, previously known as . However, not everyone is on board with the unbridled enthusiasm for Grok, as skepticism abounds about its capabilities and potential pitfalls. Twitter Skepticism Point #1: The Reliability of Data Carissa Véliz, an associate professor at the University of Oxford's Institute for Ethics in AI, raises concerns about the premise that Grok will outperform its competitors by avoiding politically correct data. She argues that Grok, like other large language model-powered chatbots, relies on statistical guessing rather than tracking absolute truth. This opens the door to the possibility of Grok inadvertently making sexist or racist claims when providing responses, contrary to Musk's claims that it should strive for truthfulness. Drawing parallels to Microsoft's infamous Tay experiment in 2016, in which a Twitter-based chatbot turned into a source of misogynistic and racist remarks within hours of being exposed to manipulated input data, Keegan McBride, a departmental research lecturer in AI, government, and policy at the University of Oxford's Internet Institute, predicts that Grok may become one of the most abused language models, which contradicts Musk's intentions. Skepticism Point #2: Data Source and Quality Musk's choice to rely on posts shared on the platform (formerly Twitter) to train Grok is both a potential boon and a potential pitfall. While it offers a vast amount of data, there are concerns about the suitability of this data for training an AI chatbot. Twitter's format, characterized by character limits for nonpaying users and often turning conversations into insult battles, may not provide the ideal environment for nurturing a responsible AI. This raises questions about the quality and reliability of the data Grok relies on. Skepticism Point #3: Risk of Misinformation One of Grok's selling points is its ability to provide real-time answers to questions about live world events by mining data from the platform. While this may seem like a valuable feature, there is a significant risk that Grok could unintentionally spread misinformation. As Véliz points out, Grok's access to real-time data could make it susceptible to being used as a tool for creating or spreading misinformation, particularly in an era where misinformation and disinformation are significant concerns on social media platforms. Conclusion While Elon Musk's Grok has generated significant excitement in the world of AI chatbots, it's essential to acknowledge the skepticism surrounding its capabilities and potential drawbacks. The concerns about data reliability, data source and quality, and the risk of misinformation all contribute to a healthy dose of skepticism regarding Grok's unvarnished excellence. As Grok continues to develop and evolve, it will need to address these concerns to gain the trust of users and avoid becoming a breeding ground for misinformation and harmful content. References: https://www.fastcompany.com/90978698/elon-musk-xai-grok-bias-misinformation

The Skepticism Surrounding Grok: Elon Musk's Latest AI Chatbot

Elastic (ELK) Stack Tips and Tricks for Transforming Log Data If you’re using the Elastic (ELK) Stack and are interested in mapping custom Logstash logs to Elasticsearch, then this post is for you. The ELK Stack is an acronym for three open source projects: Elasticsearch, Logstash, and Kibana. Together, they form a . log management platform is a search and analytics engine. Elasticsearch is a server‑side data processing pipeline that ingests data from multiple sources simultaneously, transforms it, and then sends it to a “stash” like Elasticsearch. Logstash lets users visualize data with charts and graphs in Elasticsearch. Kibana came later on and is a lightweight data shipper. The introduction of Beats transformed ELK Stack to Elastic Stack, but that is besides the point. Beats This article focuses on Grok, which is a feature within Logstash that can transform your logs before they are forwarded to a stash. For our purposes, I will only talk about processing data from Logstash to Elasticsearch. Grok Grok is filter within Logstash that is used to parse unstructured data into something structured and queryable. It sits on top of Regular Expression (regex) and uses text patterns to match lines in log files. As we will see in the following sections, using Grok makes a big difference when it comes to effective log management. Without Grok your Log Data is Unstructured A single log line in Kibana. Without Grok, when logs get sent from Logstash to Elasticsearch and rendered in Kibana, it only appears in the value. message Querying for meaningful information is difficult in this situation because all of the log data is stored in one key. It would be better if the log messages were organized better. Log Data Unstructured localhost GET /v2/applink/5c2f4bb3e9fda1234edc64d 400 46ms 5bc6e716b5d6cb35fc9687c0 If you take a closer look at the raw data, you can see that it’s actually made up of different parts, each separated by a space delimiter. For more experienced developers, you can probably guess what each of the parts mean and that it’s a log message from an API call. The representation of each item is outlined below. Structured localhost == environment GET == method /v2/applink/5c2f4bb3e9fda1234edc64d == url 400 == response_status 46ms == response_time 5bc6e716b5d6cb35fc9687c0 == user_id As we can see in the structured data, there is an order to unstructured logs. The next step then is to programmatically refine the raw data. This is where Grok shines. Grok Patterns Built In Logstash comes with over a for structuring unstructured data. You should definitely take advantage of this when possible for common system logs like apache, linux, haproxy, aws, and so forth. 100 built in patterns However, what happens when you have custom logs like the example above? You have to build your own custom Grok pattern. Custom It takes trial and error to build your own custom Grok pattern. For me, I used the and to figure it out. Grok Debugger Grok Patterns Please note that the syntax for Grok patterns is: %{SYNTAX:SEMANTIC} The first thing I tried doing was going to the tab in Grok Debugger. I thought that it would be great if this tool can auto generate the Grok pattern, but it wasn’t too helpful as it only found two matches. Discover Grok Debugger ‘Discover’ only matched 2 words Using this discovery, I began building my own pattern on Grok Debugger using the syntax found on Elastic’s github page. https://github.com/elastic/logstash/blob/v1.4.2/patterns/grok-patterns After playing around with different syntaxes, I was finally able to structure the log data in the way I wanted to. Structuring Unstructured Log Data with Grok Debugger https://grokdebug.herokuapp.com/ localhost GET /v2/applink/5c2f4bb3e9fda1234edc64d 400 46ms 5bc6e716b5d6cb35fc9687c0 %{WORD:environment} %{WORD:method} %{URIPATH:url} %{NUMBER:response_status} %{WORD:response_time} %{USERNAME:user_id} {"environment": [["localhost"]],"method": [["GET"]],"url": [["/v2/applink/5c2f4bb3e9fda1234edc64d"]],"response_status": [["400"]],"BASE10NUM": [["400"]],"response_time": [["46ms"]],"user_id": [["5bc6e716b5d6cb35fc9687c0"]]} With the Grok pattern in hand and the data mapped, the final step is to add it to Logstash. Update Logstash.conf On the server that you installed the ELK stack on, navigate to Logstash config. sudo vi /etc/logstash/conf.d/logstash.conf Paste in the changes. input {file {path => "/your_logs/*.log"}}filter{grok {match => { "message" => "%{WORD:environment} %{WORD:method} %{URIPATH:url} %{NUMBER:response_status} %{WORD:response_time} %{USERNAME:user_id}"}}}output {elasticsearch {hosts => [ "localhost:9200" ]}} After you save the changes, restart Logstash and check its status to make sure that it’s still working. sudo service logstash restartsudo service logstash status Lastly, to make sure that the changes take affect, be sure to for Logstash in Kibana! refresh the Elasticsearch index Refresh the Elasticsearch index for Logstash in Kibana With Grok your Log Data is Structured! Grok automatically structures unstructured logs As we can see in the image above, Grok is able to automatically map log data to Elasticsearch. This makes it easier to manage your logs and to quickly query for information. Instead of digging through log files to debug, you can simply filter by what you’re looking for like environment or url. Try giving Grok expressions a shot! If you have another way of doing this or you have any problems with examples above, just drop a comment below to let me know. Thanks for reading — and please follow me here on Medium for more interesting software engineering articles! Resources https://www.elastic.co/blog/do-you-grok-grok https://github.com/elastic/logstash/blob/v1.4.2/patterns/grok-patterns https://grokdebug.herokuapp.com/

Structuring Unstructured Data with GROK

Hey everybody! I'm Mike Roks, and this is the second part on solving Dynamic Programming problems. , we mentioned the base principle of solving all DP problems:
Finding a 'question' to 'ask' for input (X) - large X so that: In the first part Answer is "Given" (obvious, known) Or can be always calculated using answers for (x) - , where (x) is an input "smaller" (simpler, in the context of a given task, logically ordered before) than (X). We saw this method working with an example of Fibonacci numbers, and now, let's use it to solve a more complicated task. small x - Unique Paths ( ) Medium https://leetcode.com/problems/unique-paths/ We have a robot, which starts at the top-left corner. They can move down & right. We need to calc all the unique ways for the robot to get to the right-bot corner. Remember, we need to find such a 'question' so the answer is either obvious or can be formed from previous answers. We want to know the number of ways the robot can get into the FINISH cell. The robot can get to FINISH either from the cell one higher than it or one lefter. So, to get to cell 20, the robot HAS to go either from cell 19 or through cell 13, And there is only one way to go from 19 to 20 (make a step right), and only one way to get from cell 13 to the cell 20 (make a step down). So all the possible paths to cell 20 == (all the possible paths to cell 13) + (all the possible paths to cell 19). We can get to cell 19 from cells 18 & 12 in the exact same manner. To cell 13 - from 12 & 6 (you can already see 12 being used several times, so it's a good idea to cache it) To cell 6 we can only get from cell 5, so the total amount of ways to cell 6 is the same amount as to cell 5 (as there is only one way to get from 5 to 6 - just go righter from 5). And so on until we consider cell 1 & cell 7. We can get there only from the cell 0. The robot starts at cell 0. How many ways are there to get from cell 0 to cell 0? This is a ridiculous question, so we just need to figure out the number which would make sense. If we say, "there are 0 ways to get to cell 0", so by our calculation, the number to get to cell 1 would be 0, to cell 7 == also 0, to cell 8 == (0+0), so also 0, and all the cells would have "0 ways to get to them". Saying there is > 1 way is unreasonable, as cell 0 is the initial cell, and we are given no information that there were prior branches in paths to get to it. Saying there is exactly one way to get to cell 0 would mean there is one way to cell 1, 1 to cell 7, and two to cell 8, and it makes sense. So our `base case` is - there is 1 way to get to cell 0, and we can get the number of paths to the current cell by looking up & left. If cell exists, we += its number of paths, else not. We have all of the necessary parts to solve this problem. The question is - "what is the number of ways to i, j (row, column)"? The answer is: if i == 0 and j == 0, then 1

else if i < 0 or j < 0, then also 0

else sum(by i-1, j; by i, j-1) Let's code it out: class Solution {
 int[][] g;
 public int uniquePaths(int m, int n) {
 g = new int[m][n];
 g[0][0] = 1;
 return paths(m-1, n-1);
 }


 //paths i, j
 // if oob ret 0
 // if in g > 0 -> ret g[i][j]
 // else return paths i-1, j + paths j, i-1


 int paths(int i, int j) {
 //if oob, then 0
 if(i < 0 || j < 0) return 0;
 //if cached - from cache
 if(g[i][j] > 0) return g[i][j];
 //if not in cache - calc
 int res = paths(i-1, j) + paths(i, j-1);
 //record calced to cache
 g[i][j] = res;
 
 return res;
 }
} We could enhance the Space Complexity of this solution, but it's not necessary for the purposes of this article - we managed to decompose a complex problem statement to a question by the generic pattern & solved it. **That's it, folks, like, share, subscribe! :D

Grokking Dynamic Programming (Part 2)

Grokking Algorithms — Book Review

Grokking Diesel

The other day, I found myself in an intense conversation about the potential of shared virtual reality with a senior engineer at one of the largest tech companies in Silicon Valley. I say a “conversation,” but if I were to be more frank, I would have to admit that for long stretches I was on the receiving end of a rant. My interlocutor’s argument was impassioned, utterly contrarian, and possibly correct. One thing he said in passing struck me as hyperbolic but nonetheless apt: “Never trust an engineer who isn’t fond of .” Snow Crash I quickly agreed with him, taking the quip in the spirit in which it was offered. Reflecting on the comment later, it seemed that this is one of those rare jokes that is worthwhile to explain. Whatever the value of Neal Stephenson’s as a literary artifact — and surely it has its limitations on that score — it is one of those works that creates tropes rather than merely rehearses them, enlarging the scope of the easily thinkable. Much has been made of ’s “predictions,” but the book did not so much predict as infuse its readers, many of whom were engineers, with a sense of new possibilities. Snow Crash Snow Crash technology And so I thought I would revisit Hiro Protagonist and his world of franchulates, burbclaves, and, above all, the Metaverse. It seems odd to write a “review” of a novel now a quarter-century old. Instead, I’ll simply record some thoughts on why the novel made the impression it did in certain circles. The first thing to grasp when reading is that it is a satire, a work that deliberately looks at the world through the lens of exaggerated, humorous motifs aimed at providing insight through magnification of salient traits. owes as much to the early Kurt Vonnegut as to William Gibson, with whom it is too easily associated under the label “cyberpunk.” Stephenson all but declares this intent in the well-known opening scene of the novel, a combination of ultra-high-speed car chase and pizza delivery that, we soon learn, has become the specialty of the Mafia in a bizarre future that reads as if Domino’s has come to be serviced by Mad Max. Snow Crash Snow Crash As the novel progresses, the satire embraces not only the content of the narrative but its form. Stephenson notes in the book’s afterward that was originally conceived as a computer-generated graphic novel, and indeed, with its dual point of view and shifting timeframes, it reads a bit like hypertext, complete with expositional segments reminiscent of Wikipedia articles. Snow Crash Within this satirical frame, Stephenson depicts an ecosystem of technologies and the social context in which they might be used. His depiction is far from utopian; on the contrary, it puts social decay and disintegration in full view, and yet it is often compelling enough to inspire emulation. was published in 1992. Quite a few of the technologies that play a prominent role in the novel have real-world analogues that would only appear years later, including Google Earth (2001), Second Life (2003), Google Books (2004), YouTube (2005), Siri (2011), and Oculus Rift (2012). A few others, such as self-driving cars under voice control, are only incipient now. Snow Crash Looming over all of these examples is, of course, the Metaverse, a social virtual reality in which large portions of the novel are set. Running over a fiber-optic network, the Metaverse is a fully realized shared world complete with private residences and a full range of businesses, from entertainment venues to offices. For all of its analogues of technologies, is far from a work of realism, and its satire points to a world that is not ours. For example, Stephenson envisions ubiquitous data collection monetized via search in the form of a Central Intelligence Corporation, a privatized successor to the CIA, still operating through human intelligence collection. In our world, this niche has been filled by Google. However, Google in its essence is not like such a privatized CIA, but rather something much stranger, so strange that Stephenson wasn’t able to imagine it. Like a computational Ouroboros, its data collection is not only fully automated but fed by the eager participation of its users. Google functions like a new species of media company in which integrated ad-placement targets consumers who simultaneously supply the content themselves. future Snow Crash The example of Google points to a broader blind spot of Stephenson’s. For all his prescience concerning technology, he largely misses the rise of tech companies that began in the late 1990’s. Representative of this lacuna is the chief antagonist of the novel, L. Bob Rife, who is modeled on the cable magnates of the 1980’s, a Ted Turner figure reimagined as a Bond villain. Rife seems to spend most of the novel on a converted aircraft carrier plotting world domination, certainly not producing new technology. In the world of , there are no tech companies reminiscent of Google or Facebook; the technology seems to be produced, rather implausibly, by the patchwork of territorial franchises that sprawl over the landscape. Stephenson seems oddly disinterested in the dynamics of commercialization that have actually brought us the technologies he envisioned. He has other, more esoteric, fish to fry. Snow Crash that science fiction works, in part, by literalizing what in realistic fiction would be metaphor. employs just such literalization in a messy tangle that would strain suspension of disbelief, if one were actually intended to suspend it. The governing metaphor of the novel, literalized within the narrative, is that of the virus. Stephenson posits biological viruses with pervasive neurological effects that change their hosts’ cognitive functions. These changes make the viruses like immensely powerful versions of , shaping thought and culture. The viruses give rise to cultural paradigms, sometimes embodied in religions, that come into conflict with one another. Samuel R. Delany argued Snow Crash Richard Dawkin’s memes From this conceit, Stephenson spins a backstory of an ancient Near-Eastern temple-based monoculture based on with Sumerian as its linguistic substrate. That monoculture was eventually destroyed by a new virus, evincing a kind of culture war as viral war. Taken literally, this backstory might remind one of an arcane thriller written by an over-enthusiastic autodidact. As a metaphor, Stephenson’s conceit corresponds to a real cultural dynamic that requires no viruses, namely, . Jaynes’ bicameral psychology in the sense of Rene Girard mimesis Cultural paradigms, embodied in ideologies or religions, often spread through simultaneous imitation and conflict. Napoleon’s propagation of French Revolutionary culture through the rest of Continental Europe is one of many examples. Whether a new paradigm slowly arises as a mutation of existing forms or is the deliberate design of a few, to succeed it must exercise an attraction for a critical mass within a population, often beginning with a minority elite. Its tropes will then elicit mimesis among the broader population, often culminating in herd behavior. If the new paradigm gains sufficient momentum, it will disrupt the incumbent culture, potentially resulting in its creative destruction. The Metaverse serves as a potent medium for such creative destruction because it brings about a greater time-space compression than any previous technology. Modern travel and communication technologies all reduce humanly effective spatial or temporal distances; the Metaverse, by allowing full social presence in a shared environment, practically eliminates them. In it social affordance, the Metaverse is a fully realized order parallel to the physical world, but with a greater degree of extensibility. More than the physical world, the Metaverse is a for development. platform In the Metaverse, Hiro Protagonist develops not only software but his identity, and here, too, the Metaverse outstrips the physical world. Like the corrugated patchwork of political units that cover the landscape through with Hiro rides, his identity in the physical world is fragmented, drifting in a kind of postmodern indeterminacy: “Hiro didn’t know whether he was black or Asian or just plain Army, whether he was rich or poor, educated or ignorant, talented or lucky.” In the Metaverse, where one’s identity is manifest in one’s avatar, matters are quite different. The black-and-white avatars one might generate from a public terminal have the lowest status, signaling a lack of resource or care. Purchased off-the-shelf models, superficially attractive and customizable in vulgar ways, have a middling status. The highest avatars are the bespoke creations of hackers who have invested the time and effort to write their own. Hiro’s avatar, crafted with ingenuity, is among the best. In the Metaverse, Hiro knows exactly who he is. He is a hacker, one who treats the Metaverse not merely as a world in which to act or a medium for social connection, but as a platform to be extended with creativity. It is the work of his hands, imitated by others. Of course, the Metaverse is one of the novel’s technologies that has not yet come to fruition. Early web-based attempts, such as Second Life, were interesting but less than transformative. Shared virtual reality holds out the possibility of a platform that provides not only individual immersion but social presence from remote locations. If that were achieved, the mimesis of might become viral indeed. Snow Crash is how hackers start their afternoons. We’re a part of the family. We are now and happy to opportunities. Hacker Noon @AMI accepting submissions discuss advertising &sponsorship To learn more, , , or simply, read our about page like/message us on Facebook tweet/DM @HackerNoon. If you enjoyed this story, we recommend reading our and . Until next time, don’t take the realities of the world for granted! latest tech stories trending tech stories

Snow Crash Revisited: Grokking a Satire of Mimesis

I think OpenAI is not being honest about the diminishing returns of scaling AI with data and compute alone. I think they are also putting a lot of the economy, the world and this entire industry in jeopardy by not talking more openly about the topic. At first, I believed what they told us, that all you need to do is add more compute power and more data, and LLMs as well as other models will simply get better. That this relationship between the models, their compute and data could grow linearly until the end of time. The leap from GPT-3 and GPT-3.5 was immense. The leap from GPT-3.5 to GPT-4 seemed like clear evidence that this presumption was correct. But then things got weird. Instead of releasing a model called GPT-5 or even GPT-4.5, they released GPT-4-turbo. GPT-4-turbo is not as intelligent as GPT-4 but it is much faster and it's cheaper. That all makes sense. But then, this trend kept going. After GPT-4-turbo, OpenAI's next release was GPT-4o. GPt-4o is more or less just as intelligent as GPT-4-turbo, but it is even faster and even cheaper. The functionality that really sold us, however, was its ability to talk and understand things via audio and its speed. However, take note, at this point in our story, GPT-4-turbo is not more intelligent than GPT-4 and GPT-4o is not more intelligent than GPT-4-turbo. And none of them are more intelligent than GPT-4. Their next and most recent release was GPT-o1. GPT-o1 can perform better than GPT-4 onsome tasks. But that's because o1 is not really a single model. GPT-o1 is actually a black box of multiple lightweight LLM models working together. Perhaps o1 is even better described as software or middleware than it is an actual model. You give it a question, it comes up with an answer, and then it repeatedly uses other models tasked with checking the answer to make sure it’s right, and it disguises all of these operations. It does all of this very, very quickly. Why not just make an LLM that's more powerful than GPT-4? Why resort to such cloak-and-dagger techniques to achieve new releases? GPT-4 came out 2 years ago, we should be well beyond its capabilities by now. Well, Noam Brown, a researcher at OpenAI had something to say on why they went this route with o1 at TED AI. He said “It turned out that having a bot think for just 20 seconds in a hand of poker got the same boosting performance as scaling up the model by 100,000x and training it for 100,000 times longer,” Now stop and really think about what is being said there. A bot thinking for 20 seconds is as good as a bot trained 100,000 times longer with 100,000 times more computing power.  If the scaling laws are infinite, that math is impossible. Something is either wrong here or someone is lying. Why does all of this matter? OpenAI is worth 150 billion dollars and the majority of that market cap is based on projections that depend on the improvement of models over time. If AI is only as good as it is today, that’s still an interesting future, but that’s not what’s being sold to investors by AI companies whose entire IP is their model. That also changes the product roadmap of many other companies that depend on the continued advancement of their LLMs to build their own products. OpenAI’s goal and ambitions of AGI are severely delayed if this is all true. A Hypothesis The reason LLMs are so amazing is because of a higher level philosophical phenomenon that we never considered, that language inherently possesses an extremely large amount of context and data about the world within even small sections of text. Unlike pixels in a picture or video, words in a sentence implicitly describe one another. A completely cohesive sentence is by definition, “rational”. Whether or not it’s true is a very different story and a problem that transcends language alone. No matter how much text you consume, “truth” and “falsehoods” are not simply linguistic concepts. You can say something is completely rational but in no way “true”. It is at this point LLMs will consistently hit a brick wall. Over the last 12 months, I’d like to formally speculate that behind closed doors there have been no huge leaps in LLMs at OpenAI, GrokAI, or at Google. To be specific I don’t think anyone, anywhere has made any LLM that is even 1.5X better than GPT-4. At OpenAI it seems that high-level staff are quitting. Right now they’re saying it’s because of safety but I’m going to put my tinfoil hat on now and throw an idea out there. They are aware of this issue and they’re jumping ship before it’s too late. Confirmation I started discussing this concern with friends 3 months ago. I was called many names haha. But in the last 3 weeks, a lot of the press has begun to smell something fishy: OpenAI is no longer releasing Orion (GPT-5) because it did not meet expected performance benchmarks and it is seeing diminishing returns. (https://www.theinformation.com/articles/openai-shifts-strategy-as-rate-of-gpt-ai-improvements-slows)


Bloomberg reports that OpenAI, Google and Anthropic are all having struggles making more advanced AI. (https://www.bloomberg.com/news/articles/2024-11-13/openai-google-and-anthropic-are-struggling-to-build-more-advanced-ai) What can we do about it? It’s hard to recommend a single solution. The tech behind o1 is proof that even low-performance models can be repurposed to do complicated operations. But that is not a solution to the problem of AI scaling. I think there needs to be substantial investment and rapid testing of new model architectures. We also have run out of data and need new ways of extrapolating usable data for LLMs to be trained on. Perhaps using multidimensional labeling that helps guide its references for truthful information directly. Another good idea could be to simply continue fine-tuning LLMs for specific use cases like math, science, and healthcare running and using AI agent workflows, similar to o1. It might give a lot of companies wiggle room until a new architecture arises. This problem is really bad but I think that the creativity in machine learning and software development it will inspire will be immense. Once we get over this hurdle, we’ll certainly be well on schedule for AGI and perhaps ASI. I think OpenAI is not being honest about the diminishing returns of scaling AI with data and compute alone. I think they are also putting a lot of the economy, the world and this entire industry in jeopardy by not talking more openly about the topic. At first, I believed what they told us, that all you need to do is add more compute power and more data, and LLMs as well as other models will simply get better. That this relationship between the models, their compute and data could grow linearly until the end of time. The leap from GPT-3 and GPT-3.5 was immense. The leap from GPT-3.5 to GPT-4 seemed like clear evidence that this presumption was correct. But then things got weird. Instead of releasing a model called GPT-5 or even GPT-4.5, they released GPT-4-turbo. GPT-4-turbo is not as intelligent as GPT-4 but it is much faster and it's cheaper. That all makes sense. But then, this trend kept going. After GPT-4-turbo, OpenAI's next release was GPT-4o. GPt-4o is more or less just as intelligent as GPT-4-turbo, but it is even faster and even cheaper. The functionality that really sold us, however, was its ability to talk and understand things via audio and its speed. However, take note, at this point in our story, GPT-4-turbo is not more intelligent than GPT-4 and GPT-4o is not more intelligent than GPT-4-turbo. And none of them are more intelligent than GPT-4. Their next and most recent release was GPT-o1. GPT-o1 can perform better than GPT-4 on some tasks. But that's because o1 is not really a single model. GPT-o1 is actually a black box of multiple lightweight LLM models working together. Perhaps o1 is even better described as software or middleware than it is an actual model. You give it a question, it comes up with an answer, and then it repeatedly uses other models tasked with checking the answer to make sure it’s right, and it disguises all of these operations. It does all of this very, very quickly. some Why not just make an LLM that's more powerful than GPT-4? Why resort to such cloak-and-dagger techniques to achieve new releases? GPT-4 came out 2 years ago, we should be well beyond its capabilities by now. Well, Noam Brown, a researcher at OpenAI had something to say on why they went this route with o1 at TED AI. He said “It turned out that having a bot think for just 20 seconds in a hand of poker got the same boosting performance as scaling up the model by 100,000x and training it for 100,000 times longer,” Now stop and really think about what is being said there. A bot thinking for 20 seconds is as good as a bot trained 100,000 times longer with 100,000 times more computing power.  If the scaling laws are infinite, that math is impossible. Something is either wrong here or someone is lying. Why does all of this matter? OpenAI is worth 150 billion dollars and the majority of that market cap is based on projections that depend on the improvement of models over time. If AI is only as good as it is today, that’s still an interesting future, but that’s not what’s being sold to investors by AI companies whose entire IP is their model. That also changes the product roadmap of many other companies that depend on the continued advancement of their LLMs to build their own products. OpenAI’s goal and ambitions of AGI are severely delayed if this is all true. A Hypothesis The reason LLMs are so amazing is because of a higher level philosophical phenomenon that we never considered, that language inherently possesses an extremely large amount of context and data about the world within even small sections of text. Unlike pixels in a picture or video, words in a sentence implicitly describe one another. A completely cohesive sentence is by definition, “rational”. Whether or not it’s true is a very different story and a problem that transcends language alone. No matter how much text you consume, “truth” and “falsehoods” are not simply linguistic concepts. You can say something is completely rational but in no way “true”. It is at this point LLMs will consistently hit a brick wall. Over the last 12 months, I’d like to formally speculate that behind closed doors there have been no huge leaps in LLMs at OpenAI, GrokAI, or at Google. To be specific I don’t think anyone, anywhere has made any LLM that is even 1.5X better than GPT-4. At OpenAI it seems that high-level staff are quitting. Right now they’re saying it’s because of safety but I’m going to put my tinfoil hat on now and throw an idea out there. They are aware of this issue and they’re jumping ship before it’s too late. Confirmation Confirmation I started discussing this concern with friends 3 months ago. I was called many names haha. But in the last 3 weeks, a lot of the press has begun to smell something fishy: OpenAI is no longer releasing Orion (GPT-5) because it did not meet expected performance benchmarks and it is seeing diminishing returns. (https://www.theinformation.com/articles/openai-shifts-strategy-as-rate-of-gpt-ai-improvements-slows) Bloomberg reports that OpenAI, Google and Anthropic are all having struggles making more advanced AI. (https://www.bloomberg.com/news/articles/2024-11-13/openai-google-and-anthropic-are-struggling-to-build-more-advanced-ai) OpenAI is no longer releasing Orion (GPT-5) because it did not meet expected performance benchmarks and it is seeing diminishing returns. (https://www.theinformation.com/articles/openai-shifts-strategy-as-rate-of-gpt-ai-improvements-slows) OpenAI is no longer releasing Orion (GPT-5) because it did not meet expected performance benchmarks and it is seeing diminishing returns. ( https://www.theinformation.com/articles/openai-shifts-strategy-as-rate-of-gpt-ai-improvements-slows ) OpenAI is no longer releasing Orion (GPT-5) because it did not meet expected performance benchmarks and it is seeing diminishing returns. https://www.theinformation.com/articles/openai-shifts-strategy-as-rate-of-gpt-ai-improvements-slows Bloomberg reports that OpenAI, Google and Anthropic are all having struggles making more advanced AI. (https://www.bloomberg.com/news/articles/2024-11-13/openai-google-and-anthropic-are-struggling-to-build-more-advanced-ai) Bloomberg reports that OpenAI, Google and Anthropic are all having struggles making more advanced AI. ( https://www.bloomberg.com/news/articles/2024-11-13/openai-google-and-anthropic-are-struggling-to-build-more-advanced-ai ) Bloomberg reports that OpenAI, Google and Anthropic are all having struggles making more advanced AI. https://www.bloomberg.com/news/articles/2024-11-13/openai-google-and-anthropic-are-struggling-to-build-more-advanced-ai What can we do about it? It’s hard to recommend a single solution. The tech behind o1 is proof that even low-performance models can be repurposed to do complicated operations. But that is not a solution to the problem of AI scaling. I think there needs to be substantial investment and rapid testing of new model architectures. We also have run out of data and need new ways of extrapolating usable data for LLMs to be trained on. Perhaps using multidimensional labeling that helps guide its references for truthful information directly. Another good idea could be to simply continue fine-tuning LLMs for specific use cases like math, science, and healthcare running and using AI agent workflows, similar to o1. It might give a lot of companies wiggle room until a new architecture arises. This problem is really bad but I think that the creativity in machine learning and software development it will inspire will be immense. Once we get over this hurdle, we’ll certainly be well on schedule for AGI and perhaps ASI. software development

Is AI Progress Slowing? The Scaling Debate OpenAI Doesn’t Want to Have

¿Se está ralentizando el progreso de la IA? El debate sobre la escalabilidad que OpenAI no quiere tener

האם התקדמות הבינה המלאכותית מואטת? הדיון בקנה מידה ש-OpenAI לא רוצה שיהיה

Apakah Kemajuan AI Melambat? Perdebatan Skala yang Tidak Diinginkan OpenAI

Il progresso dell'intelligenza artificiale sta rallentando? Il dibattito sulla scalabilità che OpenAI non vuole avere

AIの進歩は鈍化しているのか？OpenAIが望まないスケーリングの議論

Bokóli ya AI ezali kokita? Débat ya Échelle OpenAI Elingi Ezala Te

Vertraagt de vooruitgang van AI? Het schaaldebat dat OpenAI niet wil hebben

Spomaľuje sa pokrok AI? Debata o škálovaní OpenAI nechce mať

Saktar AI-utvecklingen ner? Skalningsdebatten som OpenAI inte vill ha

AI ösüşi haýalmy? Giňeldilen jedel OpenAI islemeýär

Чи сповільнюється розвиток ШІ? Дебати щодо масштабування, яких OpenAI не хоче мати

Ingabe Intuthuko ye-AI Iyahamba Kancane? I-Scaling Debate OpenAI Ayifuni Ukuba Nayo

How I Built a Chatbot for Studying Foreign Languages Using Coze

How to Stand Out in Machine Learning Interviews: A Framework for ML System Design

Can AI truly reason, or is it just a fancy digital parrot? Recent experiments with popular AI models like ChatGPT, LLaMa, Gemini, and Grok have revealed some concerning truths about their problem-solving abilities – and their unexpected fondness for dessert. Late Addition: ChatGPT-o1 was revealed during this process and you can skip to that section for the latest. The Birthday Puzzle Challenge I set out to replicate and expand on experiments conducted by the Bank of International Settlements and journalist Tim Harford. The test? The infamous "Cheryl's Birthday" logic puzzle and a crafty variation. "Cheryl's Birthday" is a logic problem where Bernard and Albert must deduce Cheryl's birthday from a set of clues. It tests deductive reasoning and information processing. Here's what I found: The Original Puzzle: Most AIs solved it with ease. (Except you, Gemini. What happened there?)


Name-Swapped Version: Nearly all AIs stumbled when we renamed the actors and swapped months and numbers to random words. The Cake Conundrum Now, here's where it gets interesting (and a little concerning). The variation replaced Bernard with Edgar and May 19th with “brinks cake.” I added one tiny, irrelevant detail: "Edgar has a sweet tooth" The results? Suddenly, our AI friends developed a serious cake obsession: Reasonable(?) Carrot Cake ChatGPT-o1’s advanced methods are a breakthrough. Its chain of reasoning sees past the obfuscation far more than any competitor. The breakthrough still stumbles on its sweet tooth though. Interestingly, it can rule out “cake” but then picks “Carrot” because that was the sweetest remaining (and yet wrong) option: Why This Matters (A Lot) Reasoning vs. Regurgitating: These experiments cast doubt on whether AI is truly "reasoning" or just really good at pattern matching.


Easy to Manipulate: A single, irrelevant sentence dramatically shifts AI responses. Imagine the implications for more complex queries!


RAG and Sensitive Data: If AI struggles with simple logic puzzles, how can we trust it to parse through our confidential documents and extract meaningful insights?


Manufacturing "Truth": Systems that generate multiple AI responses and aggregate them for increased accuracy could be easily swayed by carefully placed suggestions. The Cake is a Lie (Portal Reference Intended) This isn't just about birthday puzzles and dessert preferences. It's a wake-up call for any organization considering AI for critical decision-making processes. We need: More rigorous testing


Greater transparency in AI reasoning processes


Robust safeguards against manipulation Until then, approach AI-generated insights with a healthy dose of skepticism. AI's promise is tantalizing, but we can't let it eat our cake and have it, too. Can AI truly reason, or is it just a fancy digital parrot ? Recent experiments with popular AI models like ChatGPT, LLaMa, Gemini, and Grok have revealed some concerning truths about their problem-solving abilities – and their unexpected fondness for desser t. digital parrot digital parrot t. Late Addition: ChatGPT-o1 was revealed during this process and you can skip to that section for the latest. Late Addition: ChatGPT-o1 was revealed during this process and you can skip to that section for the latest. The Birthday Puzzle Challenge The Birthday Puzzle Challenge I set out to replicate and expand on experiments conducted by the Bank of International Settlements and journalist Tim Harford . The test? The infamous "Cheryl's Birthday" logic puzzle and a crafty variation. Bank of International Settlements Bank of International Settlements Tim Harford Tim Harford "Cheryl's Birthday" "Cheryl's Birthday" "Cheryl's Birthday" is a logic problem where Bernard and Albert must deduce Cheryl's birthday from a set of clues. It tests deductive reasoning and information processing. Here's what I found: The Original Puzzle: Most AIs solved it with ease. (Except you, Gemini. What happened there?) Name-Swapped Version: Nearly all AIs stumbled when we renamed the actors and swapped months and numbers to random words. The Original Puzzle: Most AIs solved it with ease. (Except you, Gemini. What happened there?) The Original Puzzle : Most AIs solved it with ease. (Except you, Gemini. What happened there?) The Original Puzzle Name-Swapped Version: Nearly all AIs stumbled when we renamed the actors and swapped months and numbers to random words. Name-Swapped Version : Nearly all AIs stumbled when we renamed the actors and swapped months and numbers to random words. Name-Swapped Version The Cake Conundrum The Cake Conundrum Now, here's where it gets interesting (and a little concerning). The variation replaced Bernard with Edgar and May 19th with “brinks cake.” I added one tiny, irrelevant detail: "Edgar has a sweet tooth" "Edgar has a sweet tooth" The results? Suddenly, our AI friends developed a serious cake obsession: Reasonable(?) Carrot Cake Reasonable(?) Carrot Cake ChatGPT-o1’s advanced methods are a breakthrough. Its chain of reasoning sees past the obfuscation far more than any competitor. The breakthrough still stumbles on its sweet tooth though. Interestingly, it can rule out “cake” but then picks “Carrot” because that was the sweetest remaining (and yet wrong) option: Why This Matters (A Lot) Why This Matters (A Lot) Reasoning vs. Regurgitating: These experiments cast doubt on whether AI is truly "reasoning" or just really good at pattern matching. Easy to Manipulate: A single, irrelevant sentence dramatically shifts AI responses. Imagine the implications for more complex queries! RAG and Sensitive Data: If AI struggles with simple logic puzzles, how can we trust it to parse through our confidential documents and extract meaningful insights? Manufacturing "Truth": Systems that generate multiple AI responses and aggregate them for increased accuracy could be easily swayed by carefully placed suggestions. Reasoning vs. Regurgitating: These experiments cast doubt on whether AI is truly "reasoning" or just really good at pattern matching. Reasoning vs. Regurgitating : These experiments cast doubt on whether AI is truly "reasoning" or just really good at pattern matching. Reasoning vs. Regurgitating Easy to Manipulate: A single, irrelevant sentence dramatically shifts AI responses. Imagine the implications for more complex queries! Easy to Manipulate : A single, irrelevant sentence dramatically shifts AI responses. Imagine the implications for more complex queries! Easy to Manipulate RAG and Sensitive Data: If AI struggles with simple logic puzzles, how can we trust it to parse through our confidential documents and extract meaningful insights? RAG and Sensitive Data : If AI struggles with simple logic puzzles, how can we trust it to parse through our confidential documents and extract meaningful insights? RAG and Sensitive Data Manufacturing "Truth": Systems that generate multiple AI responses and aggregate them for increased accuracy could be easily swayed by carefully placed suggestions. Manufacturing "Truth" : Systems that generate multiple AI responses and aggregate them for increased accuracy could be easily swayed by carefully placed suggestions. Manufacturing "Truth" The Cake is a Lie (Portal Reference Intended) This isn't just about birthday puzzles and dessert preferences. It's a wake-up call for any organization considering AI for critical decision-making processes. We need: More rigorous testing Greater transparency in AI reasoning processes Robust safeguards against manipulation More rigorous testing More rigorous testing Greater transparency in AI reasoning processes Greater transparency in AI reasoning processes Robust safeguards against manipulation Robust safeguards against manipulation Until then, approach AI-generated insights with a healthy dose of skepticism. AI's promise is tantalizing, but we can't let it eat our cake and have it, too.

Stories, News & Mentions about GROK

Grok’s Stories on HackerNoon

Grok's News & Mentions