[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.] #  @burkov BURKOV BURKOV posts on X about ai, open ai, llm, matter the most. They currently have XXXXXX followers and 2586 posts still getting attention that total XXXXXX engagements in the last XX hours. ### Engagements: XXXXXX [#](/creator/twitter::47126544/interactions)  - X Week XXXXXXX +2% - X Month XXXXXXXXX +237% - X Months XXXXXXXXXX +9.60% - X Year XXXXXXXXXX -XX% ### Mentions: XX [#](/creator/twitter::47126544/posts_active)  - X Week XX -XX% - X Month XXX +37% - X Months XXX +33% - X Year XXXXX +55% ### Followers: XXXXXX [#](/creator/twitter::47126544/followers)  - X Week XXXXXX +1.20% - X Month XXXXXX +4.40% - X Months XXXXXX +11% - X Year XXXXXX +65% ### CreatorRank: XXXXXXX [#](/creator/twitter::47126544/influencer_rank)  ### Social Influence **Social category influence** [technology brands](/list/technology-brands) [celebrities](/list/celebrities) [social networks](/list/social-networks) [stocks](/list/stocks) [finance](/list/finance) **Social topic influence** [ai](/topic/ai), [open ai](/topic/open-ai) #154, [llm](/topic/llm) #6, [matter](/topic/matter) #795, [shocked](/topic/shocked) #741, [llms](/topic/llms) #29, [big data](/topic/big-data), [hidden](/topic/hidden), [adobe](/topic/adobe), [elements](/topic/elements) **Top accounts mentioned or mentioned by** [@linkedin](/creator/undefined) [@elonmusk](/creator/undefined) [@slowdeveloper](/creator/undefined) [@calccon](/creator/undefined) [@garymarcus](/creator/undefined) [@openai](/creator/undefined) [@jagersbergknut](/creator/undefined) [@alexshulepov7](/creator/undefined) [@ylecun](/creator/undefined) [@googleai](/creator/undefined) [@google](/creator/undefined) [@xai](/creator/undefined) [@huggingface](/creator/undefined) [@lmsysorg](/creator/undefined) [@uratioregum](/creator/undefined) [@lambdaapi](/creator/undefined) [@kimmonismus](/creator/undefined) [@microsoft](/creator/undefined) [@anthropicai](/creator/undefined) [@teknium1](/creator/undefined) **Top assets mentioned** [Alphabet Inc Class A (GOOGL)](/topic/$googl) ### Top Social Posts Top posts by engagements in the last XX hours "I said two years ago it would be Hadoop XXX and it is. When Hadoop and Big Data were at the peak of hype managers asked system administrators to connect old unused laptops that were gathering dust in the basement into a network install Hadoop on them and start "crunching data before competitors did." I didn't invent this; I witnessed it myself. No wonder XX% of Hadoop-based proofs of concept failed. With LLMs it's the same dumb managers just XX years older asking secretaries to automate business processes using Copilot. No wonder XX% of these proofs of concept failed. Companies: fire dumb" [X Link](https://x.com/burkov/status/1958187627175166384) 2025-08-20T15:21Z 50.3K followers, 39.6K engagements "This Google's paper from last year came almost unnoticed by the public but it's really an alternative architecture to the transformer that proves more parameter-efficient and effective on similar tasks. As you might know Transformers scale quadratically with sequence length. Recurrent models scale linearly but compress everything into a fixed-size hidden state that loses information over time. This paper attempts to get both: linear scaling with a memory that doesn't saturate as quickly by encoding past information into the weights of a small neural networkupdated on the fly as new tokens" [X Link](https://x.com/burkov/status/1995646053580701741) 2025-12-02T00:07Z 50.3K followers, 26.9K engagements "Adobe be like: "We have a chance to allow authors to add semantic attributes to all elements inside a PDF like this is a part of a figure and this belongs to a table." Also Adobe: "Fuck semantics Let the PDF structure be a meaningless mess"" [X Link](https://x.com/burkov/status/1997088526270132337) 2025-12-05T23:39Z 50.3K followers, 2184 engagements "This is how our galaxy really is spinning" [X Link](https://x.com/burkov/status/1996835688977998192) 2025-12-05T06:55Z 50.3K followers, 20.6K engagements "Opened my X today to see this. @elonmusk what the hell Who complained About what I received an email with a link to a post from years ago but the post was already blocked so I have no way of knowing what content was violated and whether I have to contest this strike. Given that the author of the blocked post has no way of knowing what was in the post this way one can just create DMCA complaints against anyone they don't like for some reason. Accruing multiple bogus strikes like this "may lead to suspension" of the victim's account as the below lock screen threatens. Do you think it's fair" [X Link](https://x.com/burkov/status/1997042289001271522) 2025-12-05T20:35Z 50.3K followers, 12.3K engagements "The equations that changed the world" [X Link](https://x.com/burkov/status/1987588171157098805) 2025-11-09T18:28Z 50.3K followers, 7074 engagements "Noticed that Claude is no longer printing these XX idiotic documentation files for each fix. Is it the same for you" [X Link](https://x.com/burkov/status/1995999622477586469) 2025-12-02T23:32Z 50.3K followers, 3380 engagements "@1MrPapi @DaveShapi It's the same as saying "At least we know he didn't use the spell checker." OK he didn't so what Using AI to make your text clearer isn't a sin" [X Link](https://x.com/burkov/status/1997882717246038088) 2025-12-08T04:15Z 50.3K followers, XXX engagements "The need to type "Don't code unless I tell you" when you ask a question to an LLM is extremely annoying. The person who decided to finetune an LLM to puke code on the user no matter what is a moron. And this moron works in all companies building LLMs" [X Link](https://x.com/burkov/status/1998544654866985313) 2025-12-10T00:05Z 50.3K followers, 4799 engagements "A math professor noticed his kitchen sink at home was leaking. He called a plumber. The plumber came the next day tightened a couple of nuts and the sink worked perfectly again. The professor was delighted. But when a minute later the plumber handed him the bill he was shocked. This is a third of my monthly salary Yeah I get it said the plumber. Why dont you come work for our company as a plumber Youll make three times more than you do as a professor. Just remember: when you apply say you only finished seventh grade. They dont like hiring educated people. So the professor got a job as a" [X Link](https://x.com/burkov/status/1988348230761902514) 2025-11-11T20:48Z 50.3K followers, 4.4M engagements "I use councils of models in many problem-solving tasks and have observed the same phenomenon. Models praising the answer of another model is the result of built-in sycophancy. A model takes an alternative answer as the user's pushback doesn't want to make the user angry and praises it. This is why anything multi-agent with LLMs isn't reliable. Even one agent is often wrong but when it blindly supports an even worse decision suggested by an even dumber model this is even more dangerous" [X Link](https://x.com/burkov/status/1992384681434546381) 2025-11-23T00:08Z 50.3K followers, 166K engagements "I don't understand why Google trained Gemini XXX Pro this way. User: Fix this issue. Gemini: Here's the updated code. User: You removed all the comments and log commands put them back. Gemini: Here's the code with everything back. User: The comments are still absent and the log commands too. You must only modify the relevant part of the code and keep the rest of the existing code intact. Gemini: Here's the code with everything back. User: FUCK YOU STILL REMOVED ALL THE COMMENTS AND LOGS PUT THEM BACK YOU LITTLE PIECE OF SHIT Gemini prints the correct code. @GoogleAI why I mean WHY" [X Link](https://x.com/burkov/status/1992537151821451640) 2025-11-23T10:14Z 50.3K followers, 545.2K engagements "This paper shows how to beat models with XXX billion parameters using just X million parameters on hard puzzle tasks. On the ARC-AGI reasoning benchmark this tiny model achieves XX% accuracy compared to XXXX% for DeepSeek R1 and XXXX% for OpenAI's o3-mini. On extremely difficult Sudoku puzzles with only 1000 training examples it reaches XX% accuracy where the previous best method managed 55%. The method treats problem-solving as iterative refinement. The network maintains a current answer and a separate "reasoning" state then repeatedly updates both: reason about the problem propose a better" [X Link](https://x.com/burkov/status/1992679461485994144) 2025-11-23T19:39Z 50.3K followers, 84.3K engagements "I apologize for quoting this tool. I know it hurts but it talks about something I deeply care about. You cannot steal someone's work to "cure cancer." This way every thief would claim they robbed your house to sell your stuff to cure cancer. You cannot commit a crime by saying you have good intentions. Even if the tools that didn't work a single day in their lives tell you that you can" [X Link](https://x.com/burkov/status/1995373382448165277) 2025-12-01T06:04Z 50.3K followers, 180.9K engagements "This paper really is groundbreaking. It solves a long-standing embarrassment in machine learning: despite all the hype around deep learning traditional tree-based methods (XGBoost CatBoost random forests etc) have dominated tabular datathe most common data format in real-world applicationsfor two decades. Deep learning conquered images text and games but spreadsheets remained stubbornly resistant. This paper's (published in Nature by the way) main contribution is a foundation model that finally beats tree-based methods convincingly on small-to-medium datasets and does so very fast. TabPFN in" [X Link](https://x.com/burkov/status/1996102081996861907) 2025-12-03T06:19Z 50.3K followers, 328.7K engagements "@PaulJeffries I'm sad about sklearn. There's no need to "from sklearn import" anything" [X Link](https://x.com/burkov/status/1996138506712969473) 2025-12-03T08:44Z 50.3K followers, 9707 engagements "NeurIPS 2025 Best Paper Award: Attention lets language models decide which tokens matter at each position but it has limitationsfor example a tendency to over-focus on early tokens regardless of their relevance. Gating mechanisms which selectively suppress or amplify information flow in neural networks have improved other architectures so researchers have tried adding them to attention as well. However prior attempts usually package gating together with other architectural changes making its specific contribution hard to isolate. This paper separates those effects by systematically testing" [X Link](https://x.com/burkov/status/1996682194031726784) 2025-12-04T20:45Z 50.3K followers, 77K engagements "None of the top chatbots offer the option to copy the entire conversation to the clipboard. The only option is to use cmd/ctrl-a to copy the text of the entire page which breaks the formatting and adds noise. It's the end of 2025 by the way" [X Link](https://x.com/burkov/status/1997466725206646794) 2025-12-07T00:42Z 50.3K followers, 19.2K engagements "1. LLMs as chatbots cannot reliably generate code even under an expert's supervision. X. LLMs as agents are revolutionizing every industry working as a team for hours without any human supervision. I have a feeling someone's trying to fuck me over" [X Link](https://x.com/burkov/status/1997595763333906568) 2025-12-07T09:15Z 50.3K followers, 30.6K engagements "Claude's feature to run a simple web app in the chat is so cool. If I need simplistic UI to do something like filter log lines by some keyword or regex or show various design options I just ask for it. Today I needed ideas for the icon for the new ChapterPal feature of interactive quizzes generated by the AI based on the content the user reads. So I asked for a webpage with icon options. Here's what it generated: both the separate icons and what they will look like together with the existing icons" [X Link](https://x.com/burkov/status/1998159646843146427) 2025-12-08T22:35Z 50.3K followers, 3396 engagements "The EU bureaucracy is the cancer of Europe. The still somewhat sovereign nations must start to compete again. The first step is to kick the EU's oppression machine out in disgrace" [X Link](https://x.com/burkov/status/1998168580425392564) 2025-12-08T23:11Z 50.3K followers, 5427 engagements "@scaling01 Have they already sold their equity for $6B or do they expect to sell it someday when Claude is sold to Oracle or Salesforce" [X Link](https://x.com/burkov/status/1998252673192358122) 2025-12-09T04:45Z 50.3K followers, XXX engagements "One of the most important papers of 2025 and probably of the past three years. DeepSeek R1 has removed the moat from OpenAI and given a reasoning LLM secret to humanity. R1 made the reasoning LLM the de facto standard of the modern language model. DeepSeek has shown how one can take a pretrained language model that doesn't have any practical problem-solving skills and make it learn on its own via reinforcement learning to solve the hardest scientific and coding problems. Thanks to DeepSeek we now have Claude XXX Gemini X and Kimi K2. This paper published in Nature explains how R1 was built." [X Link](https://x.com/burkov/status/1998309694268449260) 2025-12-09T08:32Z 50.3K followers, 54K engagements
[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]
@burkov BURKOV
BURKOV posts on X about ai, open ai, llm, matter the most. They currently have XXXXXX followers and 2586 posts still getting attention that total XXXXXX engagements in the last XX hours.
Engagements: XXXXXX #
- X Week XXXXXXX +2%
- X Month XXXXXXXXX +237%
- X Months XXXXXXXXXX +9.60%
- X Year XXXXXXXXXX -XX%
Mentions: XX #
- X Week XX -XX%
- X Month XXX +37%
- X Months XXX +33%
- X Year XXXXX +55%
Followers: XXXXXX #
- X Week XXXXXX +1.20%
- X Month XXXXXX +4.40%
- X Months XXXXXX +11%
- X Year XXXXXX +65%
CreatorRank: XXXXXXX #
Social Influence
Social category influence technology brands celebrities social networks stocks finance
Social topic influence ai, open ai #154, llm #6, matter #795, shocked #741, llms #29, big data, hidden, adobe, elements
Top accounts mentioned or mentioned by @linkedin @elonmusk @slowdeveloper @calccon @garymarcus @openai @jagersbergknut @alexshulepov7 @ylecun @googleai @google @xai @huggingface @lmsysorg @uratioregum @lambdaapi @kimmonismus @microsoft @anthropicai @teknium1
Top assets mentioned Alphabet Inc Class A (GOOGL)
Top Social Posts
Top posts by engagements in the last XX hours
"I said two years ago it would be Hadoop XXX and it is. When Hadoop and Big Data were at the peak of hype managers asked system administrators to connect old unused laptops that were gathering dust in the basement into a network install Hadoop on them and start "crunching data before competitors did." I didn't invent this; I witnessed it myself. No wonder XX% of Hadoop-based proofs of concept failed. With LLMs it's the same dumb managers just XX years older asking secretaries to automate business processes using Copilot. No wonder XX% of these proofs of concept failed. Companies: fire dumb"
X Link 2025-08-20T15:21Z 50.3K followers, 39.6K engagements
"This Google's paper from last year came almost unnoticed by the public but it's really an alternative architecture to the transformer that proves more parameter-efficient and effective on similar tasks. As you might know Transformers scale quadratically with sequence length. Recurrent models scale linearly but compress everything into a fixed-size hidden state that loses information over time. This paper attempts to get both: linear scaling with a memory that doesn't saturate as quickly by encoding past information into the weights of a small neural networkupdated on the fly as new tokens"
X Link 2025-12-02T00:07Z 50.3K followers, 26.9K engagements
"Adobe be like: "We have a chance to allow authors to add semantic attributes to all elements inside a PDF like this is a part of a figure and this belongs to a table." Also Adobe: "Fuck semantics Let the PDF structure be a meaningless mess""
X Link 2025-12-05T23:39Z 50.3K followers, 2184 engagements
"This is how our galaxy really is spinning"
X Link 2025-12-05T06:55Z 50.3K followers, 20.6K engagements
"Opened my X today to see this. @elonmusk what the hell Who complained About what I received an email with a link to a post from years ago but the post was already blocked so I have no way of knowing what content was violated and whether I have to contest this strike. Given that the author of the blocked post has no way of knowing what was in the post this way one can just create DMCA complaints against anyone they don't like for some reason. Accruing multiple bogus strikes like this "may lead to suspension" of the victim's account as the below lock screen threatens. Do you think it's fair"
X Link 2025-12-05T20:35Z 50.3K followers, 12.3K engagements
"The equations that changed the world"
X Link 2025-11-09T18:28Z 50.3K followers, 7074 engagements
"Noticed that Claude is no longer printing these XX idiotic documentation files for each fix. Is it the same for you"
X Link 2025-12-02T23:32Z 50.3K followers, 3380 engagements
"@1MrPapi @DaveShapi It's the same as saying "At least we know he didn't use the spell checker." OK he didn't so what Using AI to make your text clearer isn't a sin"
X Link 2025-12-08T04:15Z 50.3K followers, XXX engagements
"The need to type "Don't code unless I tell you" when you ask a question to an LLM is extremely annoying. The person who decided to finetune an LLM to puke code on the user no matter what is a moron. And this moron works in all companies building LLMs"
X Link 2025-12-10T00:05Z 50.3K followers, 4799 engagements
"A math professor noticed his kitchen sink at home was leaking. He called a plumber. The plumber came the next day tightened a couple of nuts and the sink worked perfectly again. The professor was delighted. But when a minute later the plumber handed him the bill he was shocked. This is a third of my monthly salary Yeah I get it said the plumber. Why dont you come work for our company as a plumber Youll make three times more than you do as a professor. Just remember: when you apply say you only finished seventh grade. They dont like hiring educated people. So the professor got a job as a"
X Link 2025-11-11T20:48Z 50.3K followers, 4.4M engagements
"I use councils of models in many problem-solving tasks and have observed the same phenomenon. Models praising the answer of another model is the result of built-in sycophancy. A model takes an alternative answer as the user's pushback doesn't want to make the user angry and praises it. This is why anything multi-agent with LLMs isn't reliable. Even one agent is often wrong but when it blindly supports an even worse decision suggested by an even dumber model this is even more dangerous"
X Link 2025-11-23T00:08Z 50.3K followers, 166K engagements
"I don't understand why Google trained Gemini XXX Pro this way. User: Fix this issue. Gemini: Here's the updated code. User: You removed all the comments and log commands put them back. Gemini: Here's the code with everything back. User: The comments are still absent and the log commands too. You must only modify the relevant part of the code and keep the rest of the existing code intact. Gemini: Here's the code with everything back. User: FUCK YOU STILL REMOVED ALL THE COMMENTS AND LOGS PUT THEM BACK YOU LITTLE PIECE OF SHIT Gemini prints the correct code. @GoogleAI why I mean WHY"
X Link 2025-11-23T10:14Z 50.3K followers, 545.2K engagements
"This paper shows how to beat models with XXX billion parameters using just X million parameters on hard puzzle tasks. On the ARC-AGI reasoning benchmark this tiny model achieves XX% accuracy compared to XXXX% for DeepSeek R1 and XXXX% for OpenAI's o3-mini. On extremely difficult Sudoku puzzles with only 1000 training examples it reaches XX% accuracy where the previous best method managed 55%. The method treats problem-solving as iterative refinement. The network maintains a current answer and a separate "reasoning" state then repeatedly updates both: reason about the problem propose a better"
X Link 2025-11-23T19:39Z 50.3K followers, 84.3K engagements
"I apologize for quoting this tool. I know it hurts but it talks about something I deeply care about. You cannot steal someone's work to "cure cancer." This way every thief would claim they robbed your house to sell your stuff to cure cancer. You cannot commit a crime by saying you have good intentions. Even if the tools that didn't work a single day in their lives tell you that you can"
X Link 2025-12-01T06:04Z 50.3K followers, 180.9K engagements
"This paper really is groundbreaking. It solves a long-standing embarrassment in machine learning: despite all the hype around deep learning traditional tree-based methods (XGBoost CatBoost random forests etc) have dominated tabular datathe most common data format in real-world applicationsfor two decades. Deep learning conquered images text and games but spreadsheets remained stubbornly resistant. This paper's (published in Nature by the way) main contribution is a foundation model that finally beats tree-based methods convincingly on small-to-medium datasets and does so very fast. TabPFN in"
X Link 2025-12-03T06:19Z 50.3K followers, 328.7K engagements
"@PaulJeffries I'm sad about sklearn. There's no need to "from sklearn import" anything"
X Link 2025-12-03T08:44Z 50.3K followers, 9707 engagements
"NeurIPS 2025 Best Paper Award: Attention lets language models decide which tokens matter at each position but it has limitationsfor example a tendency to over-focus on early tokens regardless of their relevance. Gating mechanisms which selectively suppress or amplify information flow in neural networks have improved other architectures so researchers have tried adding them to attention as well. However prior attempts usually package gating together with other architectural changes making its specific contribution hard to isolate. This paper separates those effects by systematically testing"
X Link 2025-12-04T20:45Z 50.3K followers, 77K engagements
"None of the top chatbots offer the option to copy the entire conversation to the clipboard. The only option is to use cmd/ctrl-a to copy the text of the entire page which breaks the formatting and adds noise. It's the end of 2025 by the way"
X Link 2025-12-07T00:42Z 50.3K followers, 19.2K engagements
"1. LLMs as chatbots cannot reliably generate code even under an expert's supervision. X. LLMs as agents are revolutionizing every industry working as a team for hours without any human supervision. I have a feeling someone's trying to fuck me over"
X Link 2025-12-07T09:15Z 50.3K followers, 30.6K engagements
"Claude's feature to run a simple web app in the chat is so cool. If I need simplistic UI to do something like filter log lines by some keyword or regex or show various design options I just ask for it. Today I needed ideas for the icon for the new ChapterPal feature of interactive quizzes generated by the AI based on the content the user reads. So I asked for a webpage with icon options. Here's what it generated: both the separate icons and what they will look like together with the existing icons"
X Link 2025-12-08T22:35Z 50.3K followers, 3396 engagements
"The EU bureaucracy is the cancer of Europe. The still somewhat sovereign nations must start to compete again. The first step is to kick the EU's oppression machine out in disgrace"
X Link 2025-12-08T23:11Z 50.3K followers, 5427 engagements
"@scaling01 Have they already sold their equity for $6B or do they expect to sell it someday when Claude is sold to Oracle or Salesforce"
X Link 2025-12-09T04:45Z 50.3K followers, XXX engagements
"One of the most important papers of 2025 and probably of the past three years. DeepSeek R1 has removed the moat from OpenAI and given a reasoning LLM secret to humanity. R1 made the reasoning LLM the de facto standard of the modern language model. DeepSeek has shown how one can take a pretrained language model that doesn't have any practical problem-solving skills and make it learn on its own via reinforcement learning to solve the hardest scientific and coding problems. Thanks to DeepSeek we now have Claude XXX Gemini X and Kimi K2. This paper published in Nature explains how R1 was built."
X Link 2025-12-09T08:32Z 50.3K followers, 54K engagements