Mame

I'm a curious technology aficionado. I enjoy sharing knowledge or interesting insights through writing articles and coding

[Jukan @jukan05]

CHINA CONSIDERS RESTRICTING OVERSEAS ACCESS TO CUTTING-EDGE AI MODELS

China’s Ministry of Commerce has led meetings over the past month with major AI companies, including Alibaba, ByteDance, and https://t.co/YDe0KRldDB, to discuss measures that would restrict overseas access to cutting-edge AI models, including models that have not yet been released.

The discussions reportedly include not only closed-source models but also open-weight models. However, the scope of application is still under debate, and the rules may ultimately apply only to future frontier models.

Officials have also discussed designating the leakage or theft of proprietary AI technologies as a national security crime, with stronger penalties, as well as restricting the types of foreign capital that can invest in Chinese AI startups.

The backdrop is the U.S. move to strengthen export controls on AI models, along with national security concerns over cutting-edge models that could possess advanced cyberattack capabilities.

Chinese authorities are reportedly concerned that advanced U.S. cybersecurity AI models could be used to exploit vulnerabilities in Chinese software.

Since the beginning of this year, China has continued to tighten measures to prevent AI technology from being transferred overseas. Authorities have investigated whether Chinese AI startups that relocated abroad violated export control laws, while also strengthening oversight of overseas transactions involving Chinese investors, technology, data, and national security concerns.

Future regulations could take the form of a tiered framework based on technological capability. Basic open-source AI models may be managed through a filing system, high-performance models may be subject to security reviews, and the most sensitive frontier models may be banned from public release or restricted to use within China.

[Amazon Web Services @awscloud]

More AI-generated code doesn't make your team faster. It might actually slow you down.

[Mame @MameliFilippo]

Benchmarks have lost most of their meaning. The only real way to evaluate these models now is to use them extensively and develop an intuition for their behavior

[Claude @claudeai]

Introducing Claude Opus 4.8: it builds on Opus 4.7 with sharper judgment, more honesty about its own progress, and the ability to work independently for longer than its predecessors.

Available today at the same price.

[Mame @MameliFilippo]

@antirez Benchmarks have lost most of their meaning. The only real way to evaluate these models now is to use them extensively and develop an intuition for their behavior

[Mame @MameliFilippo]

A chart that represents my thoughts

[Serena Ge @serenaa_ge]

Today we’re releasing DeepSWE, a new standard for agentic coding benchmarks.

On public leaderboards, top models often look relatively close in capability. DeepSWE shows where they actually diverge, reflecting the realistic experience of developers in their day-to-day work.

[Mame @MameliFilippo]

Been using Git for 10+ years and only today properly set up a global .gitignore for macOS junk files 🤦‍♂️

echo .DS_Store >> ~/.gitignore_global
git config --global core.excludesfile ~/.gitignore_global

[Mitchell Hashimoto @mitchellh]

Ghostty is leaving GitHub. I'm GitHub user 1299, joined Feb 2008. I've visited GitHub almost every single day for over 18 years. It's never been a question for me where I'd put my projects: always GitHub. I'm super sad to say this, but its time to go. https://t.co/DQDemHdytV

[Mame @MameliFilippo]

from the Os Day in Florence

[Mame @MameliFilippo]

i rawdogged a one-hour train ride doomscrolling AI hype on X

[Ben Davis @davis7]

5.5 feelings from my testing over the last few weeks:

- I hate the name. It's not a ".1" model jump, nor is it GPT-6 level, but I feel like "5.9" or something would have been more apt as it doesn't feel like an evolution of the 5 series, rather the precursor to the 6 series

- You have to use this thing on low reasoning. High is awesome, it's really powerful but 99% of tasks don't need it and it WILL over think things and do the "gpt over correct here's 10000000000000000 unit tests for your sign in button" thing. If you think u need high, try it on low first. If it can't do it, then bump up

- This thing does great with less direction and it feels like it kinda just naturally finds itself drifting towards correct solutions. U have to be really careful with prompts, agent mds, system prompts, etc. because it's sensitive as hell. If you tell it anything at any point, it will listen for good or for ill

- Still has some of that gpt style "over index on being right"-ness that can be kinda grating. It's not nearly as bad as before, but if u say the word "test" in it's zip code u better be ready for a fuck load of unit tests and "correct" code

- That said, I've also gotten this model to output some very concise and good code. One of the things I had it do was a custom virtual scroll impl in a svelte site. It perfectly used all the primitives like "bind:" and wrote an amazing implementation

- The best code I've ever seen an AI write, came from this model

- It's pretty good at writing skills for itself to use later which is cool

- Feels WAY better to talk to than 5.4 did. Still kinda has that GPT cringe, but dialed back

- I have also seen this model do some very weird stuff, and I don't think I fully know how to work with it yet. I've only done ~1 billion tokens through it so still a ton to learn

- The price sucks a ton on paper. We'll see what ends up happening. Supposedly it's way more token efficient and u should use less reasoning on it so maybe it evens out, but 2x even out? Doubt it.

Overall this is 100% my new "everything" model. It's good enough at frontend I don't really care about splitting my workflow, it's really smart, capable, and feels amazing to me. Of all the people I talked to who also had early access, I was in the minority on this one.

Time will tell how it holds up, only thing I'm certain of is that this is a very different model...

[Mame @MameliFilippo]

Of course, Deepseek v4 and GPT 5.5 had to come out the day I cannot test them

[DeepSeek @deepseek_ai]

🚀 DeepSeek-V4 Preview is officially live & open-sourced! Welcome to the era of cost-effective 1M context length.

🔹 DeepSeek-V4-Pro: 1.6T total / 49B active params. Performance rivaling the world's top closed-source models.
🔹 DeepSeek-V4-Flash: 284B total / 13B active params. Your fast, efficient, and economical choice.

Try it now at https://t.co/GCdiMzk1Dl via Expert Mode / Instant Mode. API is updated & available today!

📄 Tech Report: https://t.co/drlDrxkYtp
🤗 Open Weights: https://t.co/T13Y8i7SDM

1/n

[ARC Prize @arcprize]

GPT-5.5 on ARC-AGI (Verified)

ARC-AGI-2:
- Max: 85.0%, $1.87
- High: 83.3%, $1.45
- Med: 70.4%, $0.86
- Low: 33%, $0.35

GPT-5.5 is now state of the art on ARC-AGI-2

[Mame @MameliFilippo]

@stevibe Strange, I’m using the coding plan with no problems. Also, another thing that was causing me issues was that I was using the ‘v’ model, glm-5v, and not the glm-5.1

[Mame @MameliFilippo]

Open models must win

[Google Cloud @googlecloud]

We believe the future of AI must be open.

While other providers want to own your models, your data, and your agents, we offer you the freedom to:

🔵 choose the world’s best models
🔴 run AI wherever your data lives
🟡 control your own destiny

[left curve dev @leftcurvedev_]

how local ai feels right now

[Mame @MameliFilippo]

@thdxr I love using Handy. It has parakeet as the default model https://t.co/D98BwLIPYx

[Mame @MameliFilippo]

Shout out to Handy
I love it

https://t.co/D98BwLIPYx

[dax @thdxr]

the biggest impact on my coding workflow lately hasn't been anything agent related

it's realizing how good these parakeet local voice models are and then just dictating everything into opencode

typing is more of a burden/bottleneck than you realize