Mohamed Osman (@MohamedOsmanML) / X

Mohamed Osman

60 posts

Mohamed Osman

@MohamedOsmanML

1.01^365 =37.8 0.99^365=0.03

Calgary AB

Joined January 2016

Mohamed Osman
@MohamedOsmanML
Nov 3, 2024
We got upto 55.5% on the @arcprize leaderboard today! Progress towards the 60.2 % milestone of median human performance reported by arxiv.org/pdf/2409.01374 is not slowing down. @MindsAI_Jack @bayesilicon
14K
Mohamed Osman
@MohamedOsmanML
Aug 19, 2024
Woke up to find only a casual 3% improvement on the SOTA @arcprize today!! Huge kudos to our dream team @MindsAI_Jack @bayesilicon We're just getting started @GregKamradt @fchollet @mikeknoop
16K
Mohamed Osman
@MohamedOsmanML
Sep 6, 2024
Great to see some of our ideas reproduced. Both ARC-AGI and math proofs are extremely difficult problems where robust reasoning is crucial. This is a testament to how good of a playground ARC-AGI is! keep building!
3K
Mohamed Osman
@MohamedOsmanML
Nov 11, 2024
Really happy to win the ARC-AGI competition this year, the work on test-time methods is just getting started, lots in the works!
Jack Cole
@MindsAI_Jack
Nov 11, 2024
@arcprize 2024 with more than 16k entrants just ended after 5 months, and we rank #1 (@bayesilicon @MohamedOsmanML)! We just scored 58% with a submission that finished after the deadline! We're just getting started. We hope to have an announcement about @tufalabs soon.
968
Mohamed Osman
@MohamedOsmanML
Sep 12, 2024
ARC-AGI is where it is at! The importance and future relevance of this dataset is clear as day, that's why we have been working together at it for more than 2 years now with @MindsAI_Jack @bayesilicon and it is paying off @arcprize
Mike Knoop
@mikeknoop
Sep 11, 2024
The techniques currently on top of the ARC-AGI leaderboards are similar to those used by DeepMind's AlphaProof which got the IMO silver medal. And we still have a long way to go! The design of contest is to make the code open source and reproducible by anyone.
1.2K
Mohamed Osman
@MohamedOsmanML
Dec 9, 2024
✈️ Heading to #NeurIPS2024 today! Excited to discuss ARC, reasoning, test-time tuning, other test-time methods, anything and everything. DM me if you’re around and want to chat!
6.7K
Mohamed Osman
@MohamedOsmanML
Sep 12, 2024
Proof finding in maths (wrangling the infinitely many possible ways to frame and move forward with a mathematical problem) when done by humans has been conceptualized as this "symbolic manipulation" and "narrow" search process by some. An interesting alternate way of looking at
Mohamed Osman
@MohamedOsmanML
Sep 6, 2024
Great to see some of our ideas reproduced. Both ARC-AGI and math proofs are extremely difficult problems where robust reasoning is crucial. This is a testament to how good of a playground ARC-AGI is! keep building!
1.4K
Mohamed Osman
@MohamedOsmanML
Jul 7, 2024
Honored to be interviewed on the show, after being a fan for a really long time. It was a deep and enjoyable discussion.
Machine Learning Street Talk
@MLStreetTalk
Jun 18, 2024
We interviewed the current winners of the ARC challenge from @fchollet - Jack Cole @Jcole75Cole Mohamed Osman and Michael Hodel @bayesilicon - and commented on the new 50% result from @RedwoodResearch - just dropped on MLST
00:00
554
Mohamed Osman
@MohamedOsmanML
Sep 6, 2024
Super well deserved for such an important and very well crafted benchmark!
ARC Prize
@arcprize
Sep 5, 2024
.@fchollet recognized by @TIME as one of the 100 most influential people in AI. "With over 800 teams competing, Chollet's ARC Prize might just spark the next AI revolution from an unexpected direction." time.com/7012823/franco…
542
Mohamed Osman
@MohamedOsmanML
Sep 6, 2024
Agreed about validity of the ARC-AGI but this paper really shows that a lot of us have not come to terms with the harsh realities of IQ, and 'intelligence testing' variance and distributions. The reported average of 60% 2-shot on the ARC public-eval is quite low and surprising,
François Chollet
@fchollet
Sep 4, 2024
Replying to @fchollet
Remember -- not all humans can solve all tasks. But all tasks can be solved by humans. That's why ARC-AGI can function as a test of human-level intelligence: it saturates to 100% in the higher tranches of human intelligence.
722
Mohamed Osman
@MohamedOsmanML
Sep 6, 2024
Replying to @MohamedOsmanML
ARC-AGI is a great test of contextualization in my opinion, and definitely nowhere near solved but it may be time to reconsider how large the gap is between humans and DL-based methods in the average case.
253
Mohamed Osman
@MohamedOsmanML
Nov 11, 2024
Thanks for mentioning the team, great work!
Ekin Akyürek
@akyurekekin
Nov 11, 2024
Thanks for the attention, couple important points: 1) See @MindsAI_Jack, their team is the first one who applied method privately and they get the 1st rank in the competition. 2) See the concurrent work as well: x.com/ellisk_kellis/… 3) Obviously this is not AGI, it's a
1.1K
Mohamed Osman
@MohamedOsmanML
Nov 12, 2024
Replying to @k_nearest
Appreciate you my brother <3
26
Mohamed Osman
@MohamedOsmanML
Nov 5, 2024
Replying to @DavidRBellamy @arcprize and 2 others
haven't seen anyone try vision AFAIK, x.com/arcprize/statu… used github.com/arcprizeorg/mo… to try it with text only.
ARC Prize
@arcprize
Oct 23, 2024
Claude 3.5 Sonnet (new) scores pass@1 20.3% on 400 ARC-AGI public eval tasks. Original 3.5 Sonnet: 21%.
337