X Stats

Login Sign Up
@tonyennis

Tony Ennis

@tonyennis

4,360 Tweets
2,743 Following
2,723 Followers

Building software teams & mobile apps Prev: Built a bank in Asia YC S21

Filtered by topic: Product Commentary βœ• Clear filter
Date ↓ Tweet ❀️ πŸ”„ πŸ’¬ πŸ‘ πŸ“
Oct 10 Another paper on HN today highlighting the shortcomings of Reasoning Models. My synthesis (Ironically summarised with Claude): - Humans use deductive reasoning with state tracking - when you rule something out, it stays ruled out in your working memory. Your mental model updates and enforces consistency - ruled-out causes become hard constraints on all future reasoning. - LLMs use sequential probabilistic reasoning - they can rule something out but may "forget" and re-consider it later because they lack hard constraints. - LLMs have soft probabilistic weighting instead of explicit working memory - earlier conclusions influence but don't guarantee exclusion. - Both are "reasoning", just with different architectures and failure modes. Can LLMs ever do perfect logical reasoning? - Pure LLM architectures will likely always have limitations because they generate text token-by-token based on probabilities, not by executing logical operations - They lack explicit data structures for tracking ruled-out hypotheses or maintaining symbolic state between tokens 0 0 1 665 1.1k .
Sep 22 For years I listened to the twitter sentiment of "Don't do consumer it's just too hard". Then me & gf decided to give it a shot. Since then we've done 40m views on Tiktok and are averaging 30 new customers/day. Don't believe all the advice you hear on here. https://t.co/46GKmnelpe 9 0 1 506 259 .
Sep 15 Don't meet your (twitter) heroes folks 2 0 1 562 38 .
Nov 01 Are these new TLDs just a money making scam or what's the deal? Google launched .ing and domains are available but cost 40k/year. Seems excessive? https://t.co/jJOc21SR5d 1 0 3 660 170 .
Apr 23 I had a hunch this was the case. Still haven’t seen any examples of AutoGPT doing the crazy stuff it’s aparently capable of. Is it possible that a bunch of Twitter threadbois hyped this thing with Twitter threads and the influencers/pods ran with it (cough cough All In) 1 0 1 938 270 .