braintelligence.bsky.social - Profile | ThreadSky | a Reddit-style client for Bluesky

A next small step for reasoning models is feeding back into the model how long it’s taking to reason, and then rewarding models that reach the right answers faster

submitted 5 hours ago • 0 comments

Quantum physics paper: we spin coupled virtual fermion in phononic meta plasma with zeta entangled gluon masers Computer science paper: we asked the computer politely to be smarter and it did. We don’t know why

submitted 6 hours ago • 0 comments

Cartesia.ai distills LLaMa into a state space model with interesting results The distilled version seems to consistently outperform on reasoning tasks, even though it follows a Mamba2 architecture. So, the "reasoning" capabilities seem to emerge from the underlying architecture.

submitted 1 day ago • 1 comment

1X Technologies has launched its latest home humanoid robot, NEO Gamma. NEO can now walk with a natural human gait, swing its arms, squat down to pick up items from the ground, and sit on chairs. These dynamic control skills are learned through reinforcement learning from human motion capture data.

submitted 1 day ago • 1 comment

IBM releases a compact 2B parameter vision-language model that's "punching above its weight(s)" in visual document understanding! ✨ Paper: arxiv.org/abs/2502.09927 ✨ Model on huggingface: huggingface.co/ibm-granite/...

submitted 1 day ago • 3 comments

Side note. There are eight members of the joint chiefs. the chair is black. the CNO (navy) is a woman. the other six are white guys. chair and the cno got fired tonight.

submitted 1 day ago • 1088 comments

This is actually really amazing. Will need to dive into this when I get a chance

submitted 2 days ago • 0 comments

Prime Intellect's SYNTHETIC-1: The largest open dataset of 2M reasoning traces from DeepSeek-R1, created by compute contributors across the globe - SYNTHETIC-1: Verified math, coding and science reasoning traces - SYNTHETIC-1-SFT-7B: Fine-tuned on 800k samples www.primeintellect.ai/blog/synthet...

submitted 2 days ago • 1 comment

FYI. DeepSeek will be open-sourcing 5 repos, starting next week, sharing their small but sincere progress with full transparency. These humble building blocks in their online service have been documented, deployed and battle-tested in production.

submitted 2 days ago • 0 comments

Google Deepmind's SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features They also released model checkpoints at four sizes (ViT-B (86M), L (303M), So400m (400M), and g-opt (1B)). Repo: github.com/google-resea...

submitted 2 days ago • 1 comment

Sure looks like Google News caught the Latin Times using a large language model to write an article picking up on my latest report

submitted 2 days ago • 82 comments

I’ve been using the Stella 1.5B embedding model for a product matching system, and the results have been phenomenal, especially for such a small model. I can run it on my consumer gpu, and speed is great with PGVECTOR Highly recommend huggingface.co/NovaSearch/s...

submitted 2 days ago • 0 comments

Has anyone figured out a neural net optimizer that runs as a quantum algorithm?

submitted 2 days ago • 0 comments

After 17 years of research, Microsoft unveiled its first quantum chip using topoconductors, a new material enabling a million qubits. Current quantum computers only have dozens or hundreds of qubits. This breakthrough could revolutionize AI, cryptography, and other computation-heavy fields.

submitted 2 days ago • 5 comments

something that really is going to set realistic security thinkers apart from people stuck in the past is how cognizant one is that the US after this is going to be massively compromised. foreign and domestic security policies are going to start in equivalent of windows safe mode

submitted 3 days ago • 9 comments

UPDATE: Nearly 100 migrants, recently deported by the United States to Panama where they had been locked in a hotel, were loaded onto buses Tuesday night and moved to a detention camp on the outskirts of the jungle, several of the migrants said. www.nytimes.com/2025/02/19/w...

submitted 3 days ago • 59 comments

The iPhone 16e has arrived as expected, but I think the most important development is the introduction of Apple’s self-developed 5G baseband modem, the C1. Testing future hardware that will be widely adopted in main devices on non-mainstream models is a longstanding tradition for Apple.

submitted 3 days ago • 0 comments

Attention Scientists! Check this out! A thing you can do that's useful and likely in your wheelhouse!

submitted 4 days ago • 4 comments

I can smell the failure-stench off Jeffries right now and I’m in Oaxaca

submitted 4 days ago • 2 comments

This is a concerning milestone. Bringing new investigations and prosecutions is different than just dismissing cases against allies.

submitted 4 days ago • 94 comments

SkyReels V1 is the first open-source human-centric video foundation model. By fine-tuning HunyuanVideo on over 10 million high-quality film and television clips, it captures 33 facial expressions and over 400 natural movement combinations, accurately reflecting human emotions.

submitted 4 days ago • 1 comment

Hundreds of asylum seekers from around the globe are being trapped by the US in a Panama hotel where they'll eventually be sent to a camp in the jungle. One Chinese man said “I thought: America is a free country with respect for human rights. I had no idea it was like a dictatorship.”

submitted 4 days ago • 117 comments

Reminder, I told normies at work that RFK told a black senator blacks should receive different vaccines than whites and they just did not believe that was a real thing that happened. I know everyone on here wants things done immediately but very little news actually trickles down

submitted 5 days ago • 1 comment

Sorry, but no. Things are not “going back to normal.” This is not about downsizing govt. It’s about seizing all forms of power and wealth and trying to incapacitate physically, spiritually, politically. Our only chance is imaginative, massive, public action. Join groups, start your own.

submitted 5 days ago • 2 comments

Have to give byte dances ide a spin later www.trae.ai

submitted 5 days ago • 0 comments

🧵 No. LLMs are a tool. Like Finite Element Analysis (FEA) software to engineers. But FEA software never replaced the necessity of a formal or informal (by experience) engineering education. There was a time when the same was said about FEA software (~1996-2000).

submitted 5 days ago • 1 comment

Update: AI was completely useless for the 2nd half of this project The data file has 8192 rows AI kept insisting for hours it was db values even though they were implausibly small I plotted them in google sheets & immediately recognized it was 2 datasets one being an impulse response measurement

submitted 5 days ago • 0 comments

@mattyglesias.bsky.social says on twitter that deep research is the first ai tool to change how he works That’s a good sign

submitted 5 days ago • 0 comments

A YEAR OF SPACE FACTS: DAY 1/365 The photons we receive from the Sun take a little over 8 minutes to reach us from the surface of the Sun. But they are created deep in the core of the Sun, and it takes them 100,000s of years to get from the core to the surface!

submitted 616 days ago • 44 comments

"Law and order" in the Trump era: if you bow down to the head of the U.S. government, he may help you - whether you're awaiting trial for sex crimes in Europe or doing time in the USA for physically assaulting police.

submitted 5 days ago • 513 comments

one thing in AI is not new -- people taking one small part of a job, mischaracterizing it, ignoring all the other stuff, and then assume the AI can do the whole job on its own

submitted 5 days ago • 2 comments

And for all you Musk stans thinking "I don't like Elon either! But Mars tho!"🤡 China has a Mars mission scheduled for 2028 to return rock samples. That's before our next mission. China already has rovers on Mars. China has their own space station. And China didn't defund their own science research

submitted 5 days ago • 11 comments

NAND flash prices plunge amid supply glut, factory output cut

submitted 5 days ago • 0 comments

Update: AI was completely useless for the 2nd half of this project The data file has 8192 rows AI kept insisting for hours it was db values even though they were implausibly small I plotted them in google sheets & immediately recognized it was 2 datasets one being an impulse response measurement

submitted 5 days ago • 0 comments

I tested performance of running DeepSeek R1-14B-GGUF-Q6_K on my laptop using llama.cpp: 3.7 t/sec using --numa distributed 2.5 t/sec without numa CPU: i5-1340P - maxes out all threads RAM: 2x16GB of DDR5 using ~7GB Going to try Q5_K and Q4_K next. #buildinpublic

submitted 6 days ago • 1 comment

New LLM plugin for running models on macOS using Apple's MLX framework: llm install llm-mlx llm mlx download-model mlx-community/Llama-3.2-3B-Instruct-4bit llm -m mlx-community/Llama-3.2-3B-Instruct-4bit 'Python code to traverse a tree' simonwillison.net/2025/Feb/15/...

submitted 7 days ago • 8 comments

Another way the dictatorship might kill you. This is from a friend who is a retired air traffic controller.

submitted 6 days ago • 272 comments

Simply masking 15% of input tokens + next-token prediction (NTP) can significantly boost LLMs on key information retrieval & long-context reasoning—without extra compute! MEAP (Mask-Enhanced Autoregressive Prediction)

submitted 6 days ago • 2 comments

Intel has stated that they will separate Intel Foundry Services so that IFS can attract external investments. It seems that the plan is coming along nicely. TSMC, Broadcom, and Qualcomm will hold stakes in IFS. money.udn.com/money/story/...

submitted 6 days ago • 0 comments

Interesting use case for AI I hadn’t seen yet… the knowledge cutoff makes it like a time traveler from the past www.reddit.com/r/singularit...

submitted 6 days ago • 0 comments

The first season of solar opposites is kinda cheesy, but they are doing a good job on season 2 Might actually be better overall than Rick and Morty…

submitted 7 days ago • 0 comments