[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.]

@VictorTaelin Avatar @VictorTaelin Taelin

Taelin posts on X about agi, gpu, mindset, o3 the most. They currently have XXXXXX followers and 1060 posts still getting attention that total XXXXX engagements in the last XX hours.

Engagements: XXXXX #

Engagements Line Chart

Mentions: XX #

Mentions Line Chart

Followers: XXXXXX #

Followers Line Chart

CreatorRank: XXXXXXX #

CreatorRank Line Chart

Social Influence #

Social category influence finance technology brands countries

Social topic influence agi, gpu, mindset, o3, $neog, repo, 6969, llm, realworld, investment

Top assets mentioned Neogen Corp (NEOG)

Top Social Posts #

Top posts by engagements in the last XX hours

"sorry my verdict on Grok-4 is that it is not better than Opus for coding and not better for o3 for reasoning. I don't think it has been trained on benchmarks but I think its brain is deep friend into a problem-solution mindset that doesn't extend to real-world situations. LLMs aren't AGI so if you train your LLM to solve AIME problems it will be a great AIME problem solver. I think that's what Grok-4 is. Opus as dumb as it is (and damn is it dumb) is still the most useful model for coding today with o3 being there for the extra IQ points when you need it. I don't use anything else while we're"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-23 00:52:23 UTC 56.3K followers, 229.2K engagements

"I've chatted with one of our investors about the potential buyback offer and I want to clarify something: whenever I offer that it comes entirely from a position of worry and caution simply because I don't want someone else to be at a loss due to investing in me. Of course HOC is a startup it is a risky investment. XX% of seed-round investments go to zero. Investors know that risk and as a founder I should be in peace with that. Yet I strive to be a positive influence in the life of these who cross my path including investors. So if I can protect someone's investment at my own cost I'll often"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-26 16:19:02 UTC 56.3K followers, 11.5K engagements

"If you're really smart and want to be part of history you should probably clone NanoGPT and attempt to replicate its results without any ML method. No PyTorch no Gradients no MatMuls. Just replicate it in an entirely new creative way and you may very well end up with AGI"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-02 14:31:18 UTC 56.3K followers, 34.4K engagements

"@Pehdrew_ of LLMs - probably not it is the XX GB base model. of our AI architecture - it is the next research milestone after Bend2's launch. I have so many ideas I almost don't want to die just to watch this unfold"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-26 16:23:46 UTC 56.3K followers, XXX engagements

"crackpottery alert :3 So I think the image below could be the fundamentally canonical way to implement equality in a proof lang. Basically the identity type reduces to a value of the type it identifies. So for ex the type Id Nat X X reduces to X. We then allow values to be used a set inhabited by its own value; as in X is a value of type X. So X : Nat but also X : Set. In a way that's kinda like OTT because Id reduces but (as I've just been told) it's actually inconsistent with it since OTT Id is definitionally irrelevant whereas here coe would need to induce on the witness (and be"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-23 21:32:12 UTC 56.3K followers, 8832 engagements

"@miniapeur why mathematicians don't write empty sets in the most obvious way:"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-14 21:17:39 UTC 56.3K followers, XXX engagements

"from the names it is quite clear what they're going for: gpt-5 will be a new series of models which will replace the gpt-4o o3 dalle etc. series with an unified name. you'll still be able to select though so for example gpt-5-reasoning should be the replacement for the o-series and is likely the new name for o4 full while gpt-5-auto is a new router that auto-selects the right model for you. it is the product that roon mentioned "so that doctors aren't stuck on 4o-mini" and will be the default on ChatGPT"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-30 13:29:34 UTC 56.3K followers, XXX engagements

"Hey since Bend2's release is approaching I'll reserve this profile to talk only about it. We'll be posting under @higherordercomp's handle to start decoupling HOC from myself. We might see some rebranding because I expect this will be big. ETA is Oct 31or earlier"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-25 16:14:08 UTC 56.3K followers, 7837 engagements

"I think the most ideal outcome would be to make a mind-blowing release with Bend2 grab the attention of some of these big AI labs convince them to acquire HOC as an internal team and use a fraction of their compute to train a pure symbolic AI model. And just see how it performs. I could even rewrite NeoGen in CUDA if I had access to GPUs rather than CPUs. It is just so obviously worth to try. Why do I have to go through so many layers of indirections to do the obvious. If only humanity helped me help it"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-06 14:17:54 UTC 56.3K followers, 14.7K engagements

"there are X boxes: A B C. each box has one ball. I move all contents of A to B. I move all contents of C to A. I swap contents between B and C. I swap contents between C and A. I add X ball to each box. I add to box B the amount of balls in ball C. if no box has the same number of balls I remove one ball from each box. for each box with the same number of balls I remove one ball from it. I remove one ball from each box that has the same number of balls from another box. I add an extra box D empty. I move all contents of B to D. I move all contents of A to B. I throw all contents of D away. I"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-05 00:23:12 UTC 56.3K followers, 5045 engagements

"so here's a challenge I put you in a time chamber I give you 10000books to read for every word you read I ask you to guess the next word if you guess it right I give you a cake if you guess it wrong I zap your butt and when you're done we start over again and again and again every time you say any word that isn't in the books - I severely punish you. if you ever have a novel idea that isn't in the books - I punish you. if you ever propose a creative solution to a problem but that solution isn't in the book. I punish you. in short if you detour in any way or form from what's in the book even"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-08 02:24:09 UTC 56.3K followers, 436K engagements

"yeah it was not true guys it was trained on the answers I thought I had learned the lesson with LK-99 😭"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-24 11:53:45 UTC 56.3K followers, 56.4K engagements

"Am I crazy or is logic a non-factor in medicine As in if there is no data to support an action then that action is by default considered harmful - even when logic would say otherwise (Note: this post will be a XXX% uninformed rant about my health and topics I don't understand so it is most likely not worth a read.) For example consider these facts: X. It is well known that XX% of vzv meningitis patients report residual headache months after recovery. X. Odds of stroke are greatly increased up to several months after recovery. X. Vasculopathy post meningitis is relatively common due to"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-23 03:44:20 UTC 56.3K followers, 31.7K engagements

"so simple: you access our website you write: "pls make a clone of your fav app here" you wait a few hours you come back to a complete working web app honestly I'm too grumpy to be political today so if you don't get it sorry you're the idiot who'll be like "oooh why I didn't buy when I could it was so obvious" 🙄 of course doing so is way harder than it sounds (that's why I spent this whole year polishing NeoGen). I'm also too competent to care. I'll just build and ship it will be clear as daylight when we launch"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-01 17:20:39 UTC 56.3K followers, 26K engagements

"Some NeoGen updates So I spent the day optimizing and testing it and it found every primitive recursive function I tried instantly. Equality takes 0.0008s DrawLine is down to 0.001s and even Insert takes just 0.006s. This is several orders of magnitude faster than the previous SOTA and kinda crazy. Now the final challenge: how do we compose solutions to model complex behaviors This would not just instantly let us solve ARC-AGI but also do many other powerful things. I've implemented a first attempt today and it works: for the first time it synthesized list_to_set which requires combining 2"
@VictorTaelin Avatar @VictorTaelin on X 2025-03-26 02:48:04 UTC 56.3K followers, 25.3K engagements

"ok fuck I'm getting obsessed with this 💀 there is simply no way this is true 40%+ with XXX GPU hours. when I clear this big Bend2 task I think I'll just clone that repo and attempt to reproduce it myself. getting ready for yet another disappointment"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-22 18:19:33 UTC 56.3K followers, 94.7K engagements

"I'm still waiting for this moment TBH And that's from someone who wants to be surpassed by AIs. Mostly because all I care about is shipping great tech and using great tech. I became competent at programming studied type theory mastered the -calculus not to feel good about myself () but to get things done (probably one of the few people in the field with that mindset lol). Yet the best LLMs still fail miserably to do my job. Yesterday I spent a good X hours trying to make o3 Opus Grok etc. implement a relatively simple module on Bend2 and they just kept failing. I was tired I just wanted it"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-23 11:51:24 UTC 56.3K followers, 54.7K engagements

"@yumeacademy I get it my investors told me that several times already. I understand that. I'm fine with that (: Of course I won't always be able to back that up and I won't always offer buybacks. But sometimes I might gladly"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-27 15:16:29 UTC 56.3K followers, XXX engagements

"the startup spirit is so dull in Brazil I want to connect to ppl who want to create things with me it feels like I have to convince ppl here that doing so is fun they kinda just passively accept life this is so draining"
@VictorTaelin Avatar @VictorTaelin on X 2025-06-13 17:42:29 UTC 56.3K followers, 31.5K engagements

"BTW as some of you know my team is actively looking to join a larger AI lab so we can make this day arrive faster We're one of the few people in the world capable of building proof assistants (like Lean) compilers and GPU kernels from scratch; we built the world's fastest pure program synthesizer and there are few people alive with a deeper understanding of the -Calculus. We know how to test a model at advanced coding / theorem proving and we have many tools to improve it at that. HOC is well-funded rn but with our tiny 256Mac Mini cluster we're still far from having a fraction of a big lab's"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-23 12:33:14 UTC 56.3K followers, 37.9K engagements

"A very interesting market is emerging: hardware companies are now competing for the AI dev market with OSS models and I think they stand a very real chance against closed source labs due to the absurd speed of custom inference chips. AFAIK no major lab will have these anytime soon since they're stacked on H100's. Claude Code is still the most reliable coding agent in the world but it is SLOW. Taking 1h to perform a reasonably sized refactor greatly limits its application. I'd easily switch to a tool that was equally as intelligent but much faster. Now we have: - Groq has Kimi K2at XXX TPS. -"
@VictorTaelin Avatar @VictorTaelin on X 2025-07-25 13:57:54 UTC 56.3K followers, 12.6K engagements