fcanesin

RL the model on weight editing

One weird thing when living in the US is how hard it is to find pure grape juice - it is usually not something sold in supermarkets like orange or apple juice is. In the remainder of the west world this is not the case and normally it replaces the role cranberry juice has in the US.

Support DeepSeek and Ziphu than, I am not advocating the merits but it is the reality. ASI will be pursued regardless of wants.

Upvoted: I was going to write this exact reply. Everyone is a bit tired of hearing about AI, but it's advancing fast and it's the most important thing right now. As Demis says, if we get to ASI, we can solve all other issues with it.

HF link: https://huggingface.co/Qwen/Qwen3.8-Flash-Next

E2E can also be vibe coded and VLMs are increasingly good at it.

GLM-5.3 is further proof that all >1T models are currently undertrained. I was looking at inteligence density ( https://www.pasteboard.co/6q2-5f92mtj9.png ) from recent open models (where parameters sizes are known) and taking DS-v4-flash as upper limit GLM-5.x can 3x its performance.

Maybe was this that was the last drop for Sundar.

Demis: "I have a new amazing breakthrough"

Sundar: "Great! We really need a answer to Sol and Fable"

Demis: "They are completely owned in typhoon forecasting"

LOL https://artificialanalysis.ai/models/deepseek-v4-flash#intel...

High hopes for V4 Pro

Anthropic: reminder that DeepSeek-V4 GA version is expected to debut on July 13 as showcase for the release of the Huawei Ascend 950dt

It is not a risk is a fact - people decompiling Claude Code have found many times that it has code branchs to detect it is being used in Chinese timezone and locale.

Zhipu AI is founded by a superstar Tsinghua professor, did an IPO in January (Hong Kong stock exchange) hired half it's past research lab and it's stock is >10x since. This is not a "just distill Claude" thing.

Yes, DFlash is currently a SOTA speculative decoding method that Xiaomi just used in their MiMo model for >1000tkps

I am thinking that a small tool that simply refuses to pass large CLI output to the LLM and warns it to filter the results before reading would achieve this better as the LLM would be forced into thinking and writting the filter itself.

There are quantized processes in the brain and there is also analogic computing. So either way is just a matter of time science gets there.

Wait what!? I have been programming CUDA since 2009 and specifically remember it being pushed to C++ as main development language for the first few years, after a brief "CUDA C extension" period.

Anthropic is a great showing for startup founders how if you have a great product people will buy it, even if they dislike your pricing, your marketing and the CEO opinions.

Real PMF sells itself. The risk is of course the competition catching up, I bet switching costs are very low on this setup.

To get "End of Chat Control" EU should actually pass laws prohibiting it, this whack a mole will eventually lose.

46_255

The harness is the model "body", it's weight the cognition. Like in nature they develop together and the iteration of natural selection works at both.

If smaller labs (Zai, Moonshot, deepseek, mistral..) get together and embrace a harness, like opencode for example, as a consortium just by the power of "evolution across different environments" they might hit jackpot earlier than bigger labs.

Inserts become increasingly slow. Became >10sec for a chat completion insert after 10_000 entries on k8s Longhorn atop NVMe.

My experience trying LanceDB has been abysmal. It worked great on dev and small testing environments but as soon we tried production workloads it would get extremely slow. We shifted to PostgreSQL + pgvector and had absolutely no issues, even if it is not "engineered for multimodal data". Maybe we were doing something wrong but we did put effort in trying to make it work - it is this hard to get it performant?

Great stuff, now if could please do gemini-2.5-pro-code that would be great

Nice, congrats. But that O looks like an ass.

this, Vercel is at ~10B valuation with a business built atop React - they should and will probably take more of Meta space as stewards for it.

You are correct [although what was said at the oval office was different].

Summed together with the study visa changes: Thanks Trump for helping solve Brazil's brain drain.

yes, and it started from today.

Missing a zero here for a realistic valuation of the indisputable market leader in the most important interface of computing.

I feel like mathematicians should be able to do a second doctorate level degree a few years after their first PhD, that must be in a adjacent field of their own, but not the same.