[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.] #  @giffmana Lucas Beyer (bl16) Lucas Beyer (bl16) posts on X about llm, solve, the first, asus the most. They currently have XXXXXXX followers and 1462 posts still getting attention that total XXXXXX engagements in the last XX hours. ### Engagements: XXXXXX [#](/creator/twitter::2236047510/interactions)  - X Week XXXXXXX -XX% - X Month XXXXXXXXX +373% - X Months XXXXXXXXXX +164% - X Year XXXXXXXXXX +185% ### Mentions: XX [#](/creator/twitter::2236047510/posts_active)  - X Week XX -XX% - X Month XXX +122% - X Months XXX +138% - X Year XXX +32% ### Followers: XXXXXXX [#](/creator/twitter::2236047510/followers)  - X Week XXXXXXX +0.42% - X Month XXXXXXX +3.40% - X Months XXXXXXX +28% - X Year XXXXXXX +59% ### CreatorRank: XXXXXXX [#](/creator/twitter::2236047510/influencer_rank)  ### Social Influence **Social category influence** [technology brands](/list/technology-brands) [stocks](/list/stocks) [celebrities](/list/celebrities) [products](/list/products) [travel destinations](/list/travel-destinations) [countries](/list/countries) **Social topic influence** [llm](/topic/llm) #125, [solve](/topic/solve), [the first](/topic/the-first), [asus](/topic/asus) #651, [if you](/topic/if-you), [open ai](/topic/open-ai) #1471, [$googl](/topic/$googl), [ai](/topic/ai), [inference](/topic/inference) #53, [products](/topic/products) **Top accounts mentioned or mentioned by** [@openai](/creator/undefined) [@kolesnikov](/creator/undefined) [@xiaohuazhai](/creator/undefined) [@y0b1byte](/creator/undefined) [@asus](/creator/undefined) [@ylecun](/creator/undefined) [@jeremyphoward](/creator/undefined) [@thexeophon](/creator/undefined) [@vikhyatk](/creator/undefined) [@arohan](/creator/undefined) [@francoisfleuret](/creator/undefined) [@xjdr](/creator/undefined) [@chhillee](/creator/undefined) [@suchenzang](/creator/undefined) [@wightmanr](/creator/undefined) [@yacinemtb](/creator/undefined) [@xai](/creator/undefined) [@karpathy](/creator/undefined) [@casperhansen](/creator/undefined) [@tobyphln](/creator/undefined) **Top assets mentioned** [Alphabet Inc Class A (GOOGL)](/topic/$googl) ### Top Social Posts Top posts by engagements in the last XX hours "Tell me you never tried gpt4.5 without telling me --v" [X Link](https://x.com/giffmana/status/1996108669562290599) 2025-12-03T06:46Z 117.9K followers, 87.3K engagements "I knew *exactly* which paper she means before even opening the link. Do you" [X Link](https://x.com/giffmana/status/1996877271660118024) 2025-12-05T09:40Z 117.9K followers, 4.1M engagements "A lot of datasets are actually really bad Even big conference ones even ones that got awards It made me blanket lose trust. It's simple to find out: Just spend 30min looking at it randomly. For vision finetune a blind and a non-blind model and compare. That's all it takes" [X Link](https://x.com/giffmana/status/1994307980674658358) 2025-11-28T07:30Z 117.9K followers, 152.6K engagements "OK I have to give Jrgen this one. I've seen the Yann video like a million times but this is the first time I see the Fukushima video which is strikingly similar but was 3y earlier. Why" [X Link](https://x.com/giffmana/status/1995982071425019907) 2025-12-02T22:23Z 117.9K followers, 218.5K engagements "Honestly I'm learning more about languages I'm already familiar with by "tight loop vibe-coding" where I carefully review the vibed code refactor as we're (llm and me) writing it. For example I just learned about as_completed which I used to hand-roll but now know exists" [X Link](https://x.com/giffmana/status/1998681621961032130) 2025-12-10T09:10Z 117.9K followers, 8791 engagements "Looking at the comments it's crazy how hateful they are But i think this means people are actually scared but cannot admit it" [X Link](https://x.com/giffmana/status/1998788374984040835) 2025-12-10T16:14Z 117.9K followers, 52.2K engagements "Multimodal is unfortunately i believe much more sparsely jagged than language Language happens to have a lot of "densely jagged" regions that almost feel continuous it's what the GPT2 paper title was about ; vision not so much and audio I'm unsure. My intuition anyways" [X Link](https://x.com/giffmana/status/1993407061754102033) 2025-11-25T19:50Z 117.8K followers, 37.8K engagements "OK I'm probably gonna get some flak for this but. re the classic school(s) of RL: They are detached from reality. The whole "you can turn everything into mdp just fold stuff into the state until it's Markov. So algorithms that optimally solve mdps lead to general intelligence" might be true in infinite theory but in finite reality it makes no sense at all. That's why it's mostly ok for simple games and the music stops there with policy gradient taking over. That's why I --after some implementing playing around and thinking about both-- quickly abandoned the Bellman-school and became an adept" [X Link](https://x.com/giffmana/status/1993754340096659960) 2025-11-26T18:50Z 117.8K followers, 217.3K engagements "(long post) People previously enjoyed my laptop-posts so here's an update one year later. It's been roughly a year now that I went for the @ASUS Zenbook S14. I wanted something light quiet long battery yet powerful enough for the occasional DOTA2. Also I need linux not mac/windows; arch+i3/sway btw. This laptop fit the bill: Lunar Lake and Arc GPU turn out to be really solid and the laptop's build is gorgeous too; a lot prettier than macbooks There were initial hiccups because the CPU was too new but a month or so later it worked perfectly fine on arch and was everything I wanted. Great" [X Link](https://x.com/giffmana/status/1995227862618828890) 2025-11-30T20:26Z 117.8K followers, 97.5K engagements "@chatgpt21 Would you look at the lightbulbs please kind sir" [X Link](https://x.com/giffmana/status/1996280065970364576) 2025-12-03T18:07Z 117.8K followers, 6156 engagements "FYI you can use synthid this way if you suspect nano-banana-ing:" [X Link](https://x.com/giffmana/status/1997233910720455145) 2025-12-06T09:17Z 117.8K followers, 65.2K engagements "@unixpickle It has "cu" in the name so cRaCkEd kids will learn it instead" [X Link](https://x.com/giffmana/status/1997569081852260733) 2025-12-07T07:29Z 117.8K followers, XXX engagements "At this point I'm almost afraid to ask but. what *exactly* do people mean when they say model alignment" [X Link](https://x.com/giffmana/status/1997737169566482793) 2025-12-07T18:37Z 117.8K followers, 58.6K engagements "I can't say i would have predicted that. But with this out of the way. I think it's the same as why regressing all (or even just some) logits helps with old school classifier distillation: you force the model to learn the teachers distribution not just a point. Based on the intuition i gained in and afterwards. And here two samples are clearly telling the model "hey there's a distribution not just a 1:1 mapping". But yeah no idea how to even start finding the good threshold" [X Link](https://x.com/giffmana/status/1998151576452579550) 2025-12-08T22:03Z 117.8K followers, 4374 engagements "@Farshchi In nature especially biology everything serves a purpose No not at all plenty of purposeless things in nature ask your favorite llm for fun examples" [X Link](https://x.com/giffmana/status/1998156856930476502) 2025-12-08T22:24Z 117.8K followers, 10.7K engagements "@suchenzang Yes but this is pseudo-labeling only not function matching (= logits + mixup) which does work A LOT worse we know this nowadays:" [X Link](https://x.com/giffmana/status/1998476661721849925) 2025-12-09T19:35Z 117.8K followers, 4171 engagements "Did you do the exercise and ask your favorite frontier llm about it (or just google search but it's less fun) There are several things where we clearly know why they are there but still serve no purpose. Usually leftover from past purpose but with zero pressure to be removed. Like eyes on blind fish species for example" [X Link](https://x.com/giffmana/status/1998480721703276719) 2025-12-09T19:51Z 117.8K followers, 4088 engagements "@Deanna_Jacques You have no idea how much i used google search while raising my newborn and only because chatgpt wasn't a thing. I would have used the shit out of it" [X Link](https://x.com/giffmana/status/1998803133745819939) 2025-12-10T17:12Z 117.8K followers, 1198 engagements "Wow i XXX% agree with every single word. I though i found the perfect laptop with the Asus Zenbook S14: LunarLake actually now has good Linux support and good efficiency. Myarch bit time is something like 3-4 seconds. The build quality is great. But. it's been a little over a year and I'm now sending it into warranty for the second time. I just want a macbook air like with arch" [X Link](https://x.com/giffmana/status/1994900151161283008) 2025-11-29T22:43Z 117.9K followers, 162.7K engagements "Not saying that's what they see but if you have data and compute in the long run from scratch always wins. Here is one striking example from our distillation pape; blue is random init yellow is init with sota model *on the same target task*. And I've seen more variants of this over the years" [X Link](https://x.com/giffmana/status/1995634504321687979) 2025-12-01T23:21Z 117.9K followers, 110.6K engagements "Yo this prompt is actually pretty good I added one more line for one of my pet-peeves (see screenshot) and use it after most non-trivial changes. Interestingly I already have these in my but that seems not "strong enough". Thanks for sharing Eric" [X Link](https://x.com/giffmana/status/1996155974617268256) 2025-12-03T09:54Z 117.9K followers, 209.6K engagements "omgomgomg they are starting to go international Waymo is amazing" [X Link](https://x.com/giffmana/status/1998160639907209257) 2025-12-08T22:39Z 117.9K followers, 92.2K engagements "Hmmm Switzerland made half of Europe's university NeurIPS papers this year yeah" [X Link](https://x.com/giffmana/status/1998643435234808034) 2025-12-10T06:38Z 117.9K followers, 18.9K engagements "I have no idea about the economics or energy or heat napkin math behind this but. training models on gpus in space is pretty damn cool" [X Link](https://x.com/giffmana/status/1998863688103817646) 2025-12-10T21:13Z 117.9K followers, 9139 engagements "@_arohan_ I still ab test frontier models regularly. Opus XXX is the king at one-shotting nice onepager webapps. But for complex obscure coding questions still only Gpt5(.1) is able to solve them" [X Link](https://x.com/giffmana/status/1998998898371924427) 2025-12-11T06:10Z 117.9K followers, 3258 engagements "I am very happy to see a lot of push back on this today on my TL. Yesterday when i read it there was none and i was worried this may even be a silent majority opinion I wrote a scalding criticism but let it sit because i never rage-post. But now I'm happy to see that's no more needed. I only have one thing to add: this same person 2y ago basically wrote "well tough luck artists" regarding ai art. Well tough luck scientists" [X Link](https://x.com/giffmana/status/1999012386624225681) 2025-12-11T07:04Z 117.9K followers, 76.9K engagements "@elonmusk @wholemars @AnthropicAI Elon you realize there is no way in hell that the team would tell you "Elon the Opus is a lot better at this than Grok" right There is a yes-men bias around you that you need to correct for" [X Link](https://x.com/giffmana/status/1999024558934724787) 2025-12-11T07:52Z 117.9K followers, 18.6K engagements "@manuvector If LLMs were an academia only thing people on this panel could use connections and generally cut funding and LLM would slow down at the very least. I'm glad LLMs are not an academia only thing" [X Link](https://x.com/giffmana/status/1999031568363565090) 2025-12-11T08:20Z 117.9K followers, 2181 engagements "@JFPuget I think there was a quote from Rob Pike or some other old C guy making this statement about compilers and that eventually it's impossible to be XXX% certain the chain is not compromised" [X Link](https://x.com/giffmana/status/1999066850894974990) 2025-12-11T10:40Z 117.9K followers, 3105 engagements "Hello this is your old friend kernel svm. Yes please keep your training data around we'll need it to do inference. But don't worry only "the support vectors". Also my prediction will be interpretable since it's literally just an interpolation of the "nearest" training data support vectors How many support vectors you ask Oh Just about a third of your training set not much What the kernel means Oh it's just a kitchensink of obscure features projected in infinite dimensions kinda. Very interpret. Much demure. Wow" [X Link](https://x.com/giffmana/status/1999253373959160109) 2025-12-11T23:02Z 117.9K followers, 36.5K engagements "@SamuelMLSmith @OpenAI Nice congrats" [X Link](https://x.com/giffmana/status/1999427474997018878) 2025-12-12T10:33Z 117.9K followers, 2068 engagements "@elevenlabsio @Meta Nice" [X Link](https://x.com/giffmana/status/1999560078714601698) 2025-12-12T19:20Z 117.9K followers, 1216 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.]
@giffmana Lucas Beyer (bl16)
Lucas Beyer (bl16) posts on X about llm, solve, the first, asus the most. They currently have XXXXXXX followers and 1462 posts still getting attention that total XXXXXX engagements in the last XX hours.
Engagements: XXXXXX #
- X Week XXXXXXX -XX%
- X Month XXXXXXXXX +373%
- X Months XXXXXXXXXX +164%
- X Year XXXXXXXXXX +185%
Mentions: XX #
- X Week XX -XX%
- X Month XXX +122%
- X Months XXX +138%
- X Year XXX +32%
Followers: XXXXXXX #
- X Week XXXXXXX +0.42%
- X Month XXXXXXX +3.40%
- X Months XXXXXXX +28%
- X Year XXXXXXX +59%
CreatorRank: XXXXXXX #
Social Influence
Social category influence technology brands stocks celebrities products travel destinations countries
Social topic influence llm #125, solve, the first, asus #651, if you, open ai #1471, $googl, ai, inference #53, products
Top accounts mentioned or mentioned by @openai @kolesnikov @xiaohuazhai @y0b1byte @asus @ylecun @jeremyphoward @thexeophon @vikhyatk @arohan @francoisfleuret @xjdr @chhillee @suchenzang @wightmanr @yacinemtb @xai @karpathy @casperhansen @tobyphln
Top assets mentioned Alphabet Inc Class A (GOOGL)
Top Social Posts
Top posts by engagements in the last XX hours
"Tell me you never tried gpt4.5 without telling me --v"
X Link 2025-12-03T06:46Z 117.9K followers, 87.3K engagements
"I knew exactly which paper she means before even opening the link. Do you"
X Link 2025-12-05T09:40Z 117.9K followers, 4.1M engagements
"A lot of datasets are actually really bad Even big conference ones even ones that got awards It made me blanket lose trust. It's simple to find out: Just spend 30min looking at it randomly. For vision finetune a blind and a non-blind model and compare. That's all it takes"
X Link 2025-11-28T07:30Z 117.9K followers, 152.6K engagements
"OK I have to give Jrgen this one. I've seen the Yann video like a million times but this is the first time I see the Fukushima video which is strikingly similar but was 3y earlier. Why"
X Link 2025-12-02T22:23Z 117.9K followers, 218.5K engagements
"Honestly I'm learning more about languages I'm already familiar with by "tight loop vibe-coding" where I carefully review the vibed code refactor as we're (llm and me) writing it. For example I just learned about as_completed which I used to hand-roll but now know exists"
X Link 2025-12-10T09:10Z 117.9K followers, 8791 engagements
"Looking at the comments it's crazy how hateful they are But i think this means people are actually scared but cannot admit it"
X Link 2025-12-10T16:14Z 117.9K followers, 52.2K engagements
"Multimodal is unfortunately i believe much more sparsely jagged than language Language happens to have a lot of "densely jagged" regions that almost feel continuous it's what the GPT2 paper title was about ; vision not so much and audio I'm unsure. My intuition anyways"
X Link 2025-11-25T19:50Z 117.8K followers, 37.8K engagements
"OK I'm probably gonna get some flak for this but. re the classic school(s) of RL: They are detached from reality. The whole "you can turn everything into mdp just fold stuff into the state until it's Markov. So algorithms that optimally solve mdps lead to general intelligence" might be true in infinite theory but in finite reality it makes no sense at all. That's why it's mostly ok for simple games and the music stops there with policy gradient taking over. That's why I --after some implementing playing around and thinking about both-- quickly abandoned the Bellman-school and became an adept"
X Link 2025-11-26T18:50Z 117.8K followers, 217.3K engagements
"(long post) People previously enjoyed my laptop-posts so here's an update one year later. It's been roughly a year now that I went for the @ASUS Zenbook S14. I wanted something light quiet long battery yet powerful enough for the occasional DOTA2. Also I need linux not mac/windows; arch+i3/sway btw. This laptop fit the bill: Lunar Lake and Arc GPU turn out to be really solid and the laptop's build is gorgeous too; a lot prettier than macbooks There were initial hiccups because the CPU was too new but a month or so later it worked perfectly fine on arch and was everything I wanted. Great"
X Link 2025-11-30T20:26Z 117.8K followers, 97.5K engagements
"@chatgpt21 Would you look at the lightbulbs please kind sir"
X Link 2025-12-03T18:07Z 117.8K followers, 6156 engagements
"FYI you can use synthid this way if you suspect nano-banana-ing:"
X Link 2025-12-06T09:17Z 117.8K followers, 65.2K engagements
"@unixpickle It has "cu" in the name so cRaCkEd kids will learn it instead"
X Link 2025-12-07T07:29Z 117.8K followers, XXX engagements
"At this point I'm almost afraid to ask but. what exactly do people mean when they say model alignment"
X Link 2025-12-07T18:37Z 117.8K followers, 58.6K engagements
"I can't say i would have predicted that. But with this out of the way. I think it's the same as why regressing all (or even just some) logits helps with old school classifier distillation: you force the model to learn the teachers distribution not just a point. Based on the intuition i gained in and afterwards. And here two samples are clearly telling the model "hey there's a distribution not just a 1:1 mapping". But yeah no idea how to even start finding the good threshold"
X Link 2025-12-08T22:03Z 117.8K followers, 4374 engagements
"@Farshchi In nature especially biology everything serves a purpose No not at all plenty of purposeless things in nature ask your favorite llm for fun examples"
X Link 2025-12-08T22:24Z 117.8K followers, 10.7K engagements
"@suchenzang Yes but this is pseudo-labeling only not function matching (= logits + mixup) which does work A LOT worse we know this nowadays:"
X Link 2025-12-09T19:35Z 117.8K followers, 4171 engagements
"Did you do the exercise and ask your favorite frontier llm about it (or just google search but it's less fun) There are several things where we clearly know why they are there but still serve no purpose. Usually leftover from past purpose but with zero pressure to be removed. Like eyes on blind fish species for example"
X Link 2025-12-09T19:51Z 117.8K followers, 4088 engagements
"@Deanna_Jacques You have no idea how much i used google search while raising my newborn and only because chatgpt wasn't a thing. I would have used the shit out of it"
X Link 2025-12-10T17:12Z 117.8K followers, 1198 engagements
"Wow i XXX% agree with every single word. I though i found the perfect laptop with the Asus Zenbook S14: LunarLake actually now has good Linux support and good efficiency. Myarch bit time is something like 3-4 seconds. The build quality is great. But. it's been a little over a year and I'm now sending it into warranty for the second time. I just want a macbook air like with arch"
X Link 2025-11-29T22:43Z 117.9K followers, 162.7K engagements
"Not saying that's what they see but if you have data and compute in the long run from scratch always wins. Here is one striking example from our distillation pape; blue is random init yellow is init with sota model on the same target task. And I've seen more variants of this over the years"
X Link 2025-12-01T23:21Z 117.9K followers, 110.6K engagements
"Yo this prompt is actually pretty good I added one more line for one of my pet-peeves (see screenshot) and use it after most non-trivial changes. Interestingly I already have these in my but that seems not "strong enough". Thanks for sharing Eric"
X Link 2025-12-03T09:54Z 117.9K followers, 209.6K engagements
"omgomgomg they are starting to go international Waymo is amazing"
X Link 2025-12-08T22:39Z 117.9K followers, 92.2K engagements
"Hmmm Switzerland made half of Europe's university NeurIPS papers this year yeah"
X Link 2025-12-10T06:38Z 117.9K followers, 18.9K engagements
"I have no idea about the economics or energy or heat napkin math behind this but. training models on gpus in space is pretty damn cool"
X Link 2025-12-10T21:13Z 117.9K followers, 9139 engagements
"@arohan I still ab test frontier models regularly. Opus XXX is the king at one-shotting nice onepager webapps. But for complex obscure coding questions still only Gpt5(.1) is able to solve them"
X Link 2025-12-11T06:10Z 117.9K followers, 3258 engagements
"I am very happy to see a lot of push back on this today on my TL. Yesterday when i read it there was none and i was worried this may even be a silent majority opinion I wrote a scalding criticism but let it sit because i never rage-post. But now I'm happy to see that's no more needed. I only have one thing to add: this same person 2y ago basically wrote "well tough luck artists" regarding ai art. Well tough luck scientists"
X Link 2025-12-11T07:04Z 117.9K followers, 76.9K engagements
"@elonmusk @wholemars @AnthropicAI Elon you realize there is no way in hell that the team would tell you "Elon the Opus is a lot better at this than Grok" right There is a yes-men bias around you that you need to correct for"
X Link 2025-12-11T07:52Z 117.9K followers, 18.6K engagements
"@manuvector If LLMs were an academia only thing people on this panel could use connections and generally cut funding and LLM would slow down at the very least. I'm glad LLMs are not an academia only thing"
X Link 2025-12-11T08:20Z 117.9K followers, 2181 engagements
"@JFPuget I think there was a quote from Rob Pike or some other old C guy making this statement about compilers and that eventually it's impossible to be XXX% certain the chain is not compromised"
X Link 2025-12-11T10:40Z 117.9K followers, 3105 engagements
"Hello this is your old friend kernel svm. Yes please keep your training data around we'll need it to do inference. But don't worry only "the support vectors". Also my prediction will be interpretable since it's literally just an interpolation of the "nearest" training data support vectors How many support vectors you ask Oh Just about a third of your training set not much What the kernel means Oh it's just a kitchensink of obscure features projected in infinite dimensions kinda. Very interpret. Much demure. Wow"
X Link 2025-12-11T23:02Z 117.9K followers, 36.5K engagements
"@SamuelMLSmith @OpenAI Nice congrats"
X Link 2025-12-12T10:33Z 117.9K followers, 2068 engagements
"@elevenlabsio @Meta Nice"
X Link 2025-12-12T19:20Z 117.9K followers, 1216 engagements