Hacker Newsnew | past | comments | ask | show | jobs | submit | storus's commentslogin

Wouldn't this model prevent governments from installing and keeping backdoors alive? One could just audit their whole software stack with it and get super resilient to any attack which might not play nicely with the people in power that want some backdoors open. I would think that's one of the main reasons to keep the model non-public.

Only thanks to Meta we have competitive local LLMs. Without LLama nothing decent would have been released. Commoditize your complements in action.

Does this mean that Iran will have functional nukes in two weeks? Given how previous "ceasefires" turned out (blowing up their leadership), I don't think they are naive again and don't seem desperate to end it.

No, there is more to building a functioning nuke than just fuel enrichment.

The access to highly enriched uranium or separated plutonium is the limiting factor for construction of nuclear weapons.

It really depends on how small and how efficient you need to make weapon, a nuclear weapon fitting inside a rocket nose cone is much more sophisticated than nuclear weapon that has to be only transportable by truck, ship or airplane.

For example, the simple design of Little Boy used on Hiroshima contained 64 kilograms of uranium, but less than a kilogram underwent nuclear fission.

https://en.wikipedia.org/wiki/Little_boy

"Unlike the implosion design developed for the Trinity test and the Fat Man bomb design that was used against Nagasaki, which required sophisticated coordination of shaped explosive charges, the simpler but inefficient gun-type design was considered almost certain to work, and was never tested prior to its use at Hiroshima."

So with access to highly enriched uranium (enrichment greater than 90%), a large and crude bomb could be produced in few weeks. How could they deliver it anywhere? They don't have airplanes. Truck? Speedboat?


Given how the past nuclear deals went over decades, there's little hope of follow through now.

JCPoA compliance was verified by the US and the IAEA regularly until the agreement was suspended by Donald Trump in 2018.

And the agreement continued with other nations, but IAEA started raising concerns in 2019 and Iran started breaking conditions. A bunch of countries tried to reinstate it under Biden but the Iranians wouldn't do it. Maybe they would have stayed compliant, maybe not. We'll never know. What we do know is that they wouldn't continue the agreement with other parties nor recommit to it later.

I am wondering when Dr. Phoxotic makes it to the top 10...

That's cool! You can add SoulX-FlashHead for real-time AI head animation as well if you want to simulate a teacher.

Thanks for sharing! I'm still torn about it. Sure it'll feel more natural if you have the AI head animation, but I don't want people to get attached to it. I don't want to make the loneliness epidemic even worse.

I am forced to use Yandex just to find stuff Google or DDG never shows anymore.

Interesting. Do you have some examples?

Cable TV->enshittification->YouTube to the rescue->enshittification->???

Yeah, Qwen3 coder for Claude Code and 3.5 for OpenClaw replaced my full-stack use of Opus 4.6 already; it's fine for basic web apps, k8s/docker infra setup, optimizing AI models etc. with only slightly higher error rate than Opus. Upcoming 3.6 together with Gemma4 might make it even better (still to test). OpenAI's memory spot market play might have been directed at local inference as well.

Look for Deepseek 4 when it drops, I’m curious how good it will be.

The thing is, if you’re using AI responsibly today you’re already breaking down tasks to such a granular level that you don’t need the power of Opus. You can save that for deeper research tasks.


Based on the current DeepSeek website I suspect it's not going to be great as their current model (V3.4? V4-mini?) often forgets or changes facts explicitly mentioned in the conversation which R1 never did. It's better than R1 at math or coding, but nearly unusable for deep conversation. I suspect they pushed MLA or linear attention too much, or quantize a lot more than before.

I keep paying ElevenLabs for 3 years after some early AI agent project where I used it as my payment data is bound to a google account associated with a phone number that expired in the meantime. I thought adding the google authenticator to the account and switching to it as a secondary authentication method from the phone number would allow me to cancel the subscription, but for some reason Google insists I verify using the expired phone number...

Can't you use Claude caveman mode?

https://github.com/JuliusBrussee/caveman


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: