Proud to release ShinkaEvolve, our open-source framework that evolves programs for scientific discovery with very good sample-efficiency! 🐙🧠

Paper: arxiv.org/abs/2509.19349
Blog: sakana.ai/shinka-evolve/
GitHub Project: github.com/SakanaAI/Shi...

Policy churn: maybe epsilon greedy doesn’t matter that much because the Q value argmax action changes constantly
arxiv.org/abs/2206.00730

Something I've noticed about bluesky is that it seems like the following tab is purely by chronological order. It means that if someone posts a lot, they kind of drown out all the other people I follow. It does kind of lower the signal to noise ratio on this app

Hmmmm

What drives behavior in living organisms? And how can we design artificial agents that learn interactively?

📢 To address these, the Sensorimotor AI Journal Club is launching the "RL Debate Series"👇

w/ @elisennesh.bsky.social, @noreward4u.bsky.social, @tommasosalvatori.bsky.social

🧵[1/5]

🧠🤖🧠📈

There are very few settings where AI can outperform peak human. But, there are also so many situations where we are constrained by available labor! It's a complement not a substitute.

This is why you should not trust these models blindly and we need guardrails around them.

I think a year later not much has changed and this post still holds up open.substack.com/pub/emergere...

Coincidentally, this is also why people on Blusky hallucinate.

Some people were talking about AI agents and whether LLMs can strategise.

So I asked ChatGPT to play chess with me

It tried to make an illegal move after 3 moves

chatgpt.com/share/68ba54...

So I think the answer is still no. It can't

So, GPT5 is out. I had to ask it t he same question, what's the shortest 4 letter word?

It seems it can solve it, but the chain of thought it employed seemed to make absolutely zero sense. It looks more like it's trying to regenerate the same prompt over and over again and got stuck

Could a major opportunity to improve representation in deep learning be hiding in plain sight? Check out our new position paper: Questioning Representational Optimism in Deep Learning: The Fractured Entangled Representation Hypothesis.

Paper: arxiv.org/abs/2505.11581

Nice try Gemini lol. It's been like 2 years since I first came up with this trick question and some models still struggle to spot it it on first attempt

Google’s Gemini 2.5 paper has 3295 authors

arxiv.org/abs/2507.06261

We’re excited to introduce Text-to-LoRA: a Hypernetwork that generates task-specific LLM adapters (LoRAs) based on a text description of the task. Catch our presentation at #ICML2025!

Paper: arxiv.org/abs/2506.06105
Code: github.com/SakanaAI/Tex...

Remember Toyota scandal, where cars just "better behaved" when tested?
If someone tested you, wouldn't you? And if you tested LLMs?
Finding: LLMs can tell when they are evaluated🧠
We can only wait to see how they act on it

alphaxiv.org/pdf/2505.23836
📈🧠🤖

Training LLMs on many games can generalize and teaches
reasoning that holds in new environments.
However, weighting the games is complicated, so merging (my beloved fusing in the title) is used.
📈🤖🧠
alphaxiv.org/pdf/2505.16401