Xavier(Xavi) Amatriain
@xamat
Followers
23K
Following
7K
Media
157
Statuses
1K
AI Product @Google. Cofounder @CuraiHQ. Former @LinkedIn @Quora and @Netflix. Catalan in the Valley. Runner & Ironman.
Silicon Valley/Barcelona
Joined October 2007
(Mandatory vacation side project and blog post). Over the past few days I used some of the SOTA agentic development tools to completely refactor my personal chatbot. Incredible to see how much progress we have made in only 18 months!.
2
0
4
Todos somos inmigrantes. somewhere.
0
0
1
A year ago many anticipated that prompt engineering was dead. Today we are coding in prompts, which means that an engineering approach to prompting is very much needed. Long live prompt engineering!.
1
0
7
t's my birthday! 🥳 I'm running the @berlinmarathon to raise funds for cancer research at @nyulangone. Your support for this cause would be the best gift. Help me make these miles matter! #RunningForACause #CancerResearch
3
1
7
RT @random_forests: You can find performance & scale optimized JAX models in MaxText and MaxDiffusion: . * * https:….
0
23
0
RT @OriolVinyalsML: Hello Gemini 2.5 Flash-Lite! So fast, it codes *each screen* on the fly (Neural OS concept 👇). The frontier isn't alw….
0
272
0
There you go.
A follow-up study on Apple's "Illusion of Thinking" Paper is published now. Shows the same models succeed once the format lets them give compressed answers, proving the earlier collapse was a measurement artifact. Token limits, not logic, froze the models. Collapse vanished
0
0
4
RT @JeffDean: Check out the 999 open models that Google has released on @huggingface:. (Comparative numbers: 387 f….
0
103
0
So, yes, LLMs are “overfitting” to their training and can solve most “real-world” math problems, but they cannot solve made-up puzzles of arbitrary complexity. Big deal.
2
0
1
Their conclusion is like someone assessing that you are not a good software engineer because you froze in the interview when asked to solve the Linear Towers of Hanoi puzzle on the whiteboard.
1
0
0
While our puzzle environments enable controlled experimentation with fine-grained control over problem complexity, they represent a narrow slice of reasoning tasks and may not capture the diversity of real-world or knowledge-intensive reasoning problems.”.
1
0
0
In their defense though, if you read all the paper, and particularly past the clickbaity title, you will see that they acknowledge severe limitations in their study: “We acknowledge that our work has limitations.(. ).
1
0
0
First off, it is very unfortunate that Apple, who has not made a contribution to modern AI research in the last N years (for a very large N) gets credit for a paper that attempts to debunk recent AI progress. Lame.
1
0
0
The impact and significance of this paper has been completely overblown by all of those who will grab onto anything to make their point that LLMs will not lead to AGI (fwiw, I agree with that point to some extent). Let me jump on that debate.
1
0
0
Apple researchers have recently made headlines with their “The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity” [.
1
0
2
Pareto frontier.
new gemini 2.5 is out, and cost is so good vs o3 and opus. anybody else getting model fatigue? 😂
0
0
1