Jon Ander Beracoechea (@jonberako) header
Jon Ander Beracoechea (@jonberako) profile picture

Posting mainly about AI. Avid board gamer and lazy runner & cook. Dad of 2 girls. Chief Scientist @BBVA. Opinions are my own

Madrid
Joined August 2023
Photo by Jon Ander Beracoechea (@jonberako)Photo by Jon Ander Beracoechea (@jonberako)Photo by Jon Ander Beracoechea (@jonberako)Photo by Jon Ander Beracoechea (@jonberako)Photo by Jon Ander Beracoechea (@jonberako)Photo by Jon Ander Beracoechea (@jonberako)Photo by Jon Ander Beracoechea (@jonberako)Photo by Jon Ander Beracoechea (@jonberako)
Jon Ander Beracoechea retweeted
(Spanish) Un placer compartir esta conversación sobre la posibilidad de emoción en IAs con Ángel Rodríguez. Gracias por invitarme a este programa!
¿Pueden las máquinas aprender a sentir emociones? Alberto Hernández nos presenta un innovador marco de autoaprendizaje emocional para la IA. Podcast "Hablando con Científicos. #hablandoconcientificos cienciaes.com/entrevistas/20… @cienciaes
2
4
206
Jon Ander Beracoechea retweeted
"As with computer enthusiasts in 2004, mainstream analysts and journalists are missing the forest for the trees: despite the slowing down of one trend, the industry collectively remains moving forward at a breakneck pace due to other new emerging paradigms that are ripe for scaling and expansion. It is possible to stack “scaling laws” – pre-training will become just one of the vectors of improvement, and the aggregate “scaling law” will continue scaling just like Moore’s Law has over last 50+ years." semianalysis.com/2024/12/11/…
4
3
35
1,926
Jon Ander Beracoechea retweeted
Extremely true from @tylercowen. Not just for books, but for almost everything. Music, movies, Netflix documentaries, essays, even Twitter threads, codebases and strategy docs, PRDs, the marginal benefit to asking more questions has increased dramatically even as the cost fell.
17
26
6
275
26,114
Jon Ander Beracoechea retweeted
Hard to find people who simultaneously 'get' language models deeply without getting lost in them, have a good understanding of human behaviours, don't approach every sociopolitical question as binaries, understand systemic dynamics and complexity without robbing individuals of agency, hold opposing ideas in mind while retaining the ability to function, and can both zoom in and out of all the above without overindexing on any particular aspect.
49
24
4
373
21,398
Jon Ander Beracoechea retweeted
Not groundbreaking insight here but I think the skills that will matter the most in the coming decade is being adaptive, quickly grasping new developments, being highly agentic, being good at specifying intent in detail, and using tools/AI early and well.
15
13
2
131
5,406
Jon Ander Beracoechea retweeted
Two things that both evolution and neural networks share in common is they both thrive on scale, and they were both once dismissed as obsolete to AI. The deep evolution moment is awaiting.
Replying to @fchollet
Similarly, there are several research directions today seen as long-abandoned failures, that are only waiting for the right amount of attention and the right level of compute scale to shine. Genetic algorithms are one of them.
6
16
124
28,068
Jon Ander Beracoechea retweeted
There are two ways AI impacts organizations: 1) The usual way, which will take a decade to play out as organizations adjust to AI & integrate into systems 2) Via apotheosis, where AGI creates a parallel organizational ecosystem instantly Most people betting on #1. Labs bet on #2
26
35
6
276
26,493
Jon Ander Beracoechea retweeted
Introducing, Act-One. A new way to generate expressive character performances inside Gen-3 Alpha using a single driving video and character image. No motion capture or rigging required. Learn more about Act-One below. (1/7)
298
1,130
808
6,861
1,140,656
Jon Ander Beracoechea retweeted
We now have the full connectome of the fruit fly. It has 140K neurons and 50M synapses. You can explore the connectome in 3D with an online tool. This is auditory circuitry.
6
14
5
179
24,126
Jon Ander Beracoechea retweeted
Hot take 🔥: I think there is a huge overinflated perception of how interested consumers are in having AI Agents do shopping / browsing tasks for them. People generally want agents that will do the boring parts for them, but many of those tasks tend to be higher risk...
125
62
43
1,429
372,618
Jon Ander Beracoechea retweeted
Recent results like Apple’s show that LLMs (even o1) flub on reasoning with simple changes to problems that shouldn’t matter. A consensus is building that it shows they are “just pattern matching.” But that metaphor is misleading: good reasoning itself can also be framed as “just pattern matching” at each step. The issue is not that we are merely seeing pattern matching, but that we are seeing *bad* pattern matching, at the wrong level of abstraction. If you think about it, that is a more serious pathology because it doesn’t separate when it works vs. when it doesn’t work into conveniently distinct buckets of computational tasks, In a sense, calling it “just pattern matching” implies an easier fix than there really is, as if all it will take is a better o1.
28
24
4
210
36,729
Jon Ander Beracoechea retweeted
The future is coming fast, and will be much stranger than we can imagine.
1
1
9
300
Jon Ander Beracoechea retweeted
Replying to @tsarnick
If intelligence is just 'competence at playing games,' then I’ve already been outsmarted—since I regularly lose solitaire board games to a glorified flowchart...😬
1
1
131
Jon Ander Beracoechea retweeted
AI company’s training and compute costs are enormous. You need $5B/year to stay in the race.
6
12
56
9,557
Jon Ander Beracoechea retweeted
Update: A couple of us Google Brain / DeepMind researchers have left to work on something new. Some observations: * Off-the shelf AI libraries (e.g. RAG and agents) do not work well enough for high-quality products. * In-house research is essential to get the higher quality/reliability that transforms a cool, but flaky demo into an indispensable tool. * Recent competitive trends and a paradigm shift in scaling compute is drying up the moat around traditional foundation model companies. * There are hints of AI disruption in some fields, but it is still early innings and is inevitable for most knowledge work. Finally, a cracked team can get a lot more done with far fewer people these days.
78
201
38
2,773
294,467
Jon Ander Beracoechea retweeted
1/ New data insight! How much AI compute is out there, and who owns it? We estimate the total number of AI chips that NVIDIA has sold since early 2022, as well as the total number of TPUs that Google has deployed. 🧵
11
79
30
482
495,955
Jon Ander Beracoechea retweeted
Who’s writing the substack post about how winning back to back nobel prizes is a sign AI is hitting a wall?
1
2
6
650
Jon Ander Beracoechea retweeted
The latest Lex Fridman podcast with a GREAT group of AI nerds (who had some substantial stuff to say about the current state and future of development) ends with the quote: "Nothing is as permanent as a temporary solution that works." Spoken like a true indie hacker :D
13
7
2
172
14,084
Jon Ander Beracoechea retweeted
Has anyone tried feeding the transcript of a NotebookLM-generated podcast back to NotebookLM to generate a new podcast? And then iterate? Do we get a fixed-point, periodic, or chaotic attractor?
29
41
9
582
95,900