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

It's relevant because he's the CEO of one of the largest US companies that makes export-restricted, highly-coveted tech with military applications.

Because it's a matter of national security.

And because he deals with foreign governmennts and should know better being of Chinese origin that the Chinese are known to use such honeypots?


You seriously think export controls actually prevent China from getting GPUs in a variety of ways?

Remember that the CIA was able to get Titanium from the USSR to build the SR 71 during the hight of the Cold War. And the USSR was more locked down than the US is with export controls.


With Strata and specialized engines, Qwen Flash Next runs even on 16GB cards.

I use it at 60+ tok/second on a single 24GB card.


They might have spent next to 0 time actually 'hacking' - they may have just told Opencode to hack Opencode website and let it run for hours or days till it succeeded.


Hm? Jev's actual (network) latency is not that great and even small local models are doing far better latency-wise.

The Microsoft article GP linked even shows the MS model having 95ms latency.

Even on price Jev being matched (the same MS model is "Input tokens cost $0.042 USD per million tokens. Output tokens are free.", same as Jev).

Cloudflare's Clef-flash model is actually slightly cheaper: "$0.038 in / $0 out per 1M" too.

Jev is being matched or exceeded in performance and price within a month of them going public.


Coder is not as good and seems to have noticable loss.

Use the normal one - I've used that for long coding sessions and it worked well on multi-hour runs.


@the_mitsuhiko - How is the session replay of Pi embedded in the article?

It doesn't look to be asciinema?


> @the_mitsuhiko - How is the session replay of Pi embedded in the article?

Some bespoke slop :)


I'm seeing 128GB DDR5 RAM costing $3500+ just by itself.

How are you going to build an entire system shipped for $3k?

This spec sheet looks more like a dreamed-up spec sheet than pricing reality.


Is the Magic 8 ball immune to hallucination as Jev is?


Yes. It's exactly as immune to hallucination - it can only produce answers from a small fixed set.


It would be good to list 1) zero-shot accuracy and 2) latency on the models page . The LLM-based models' latency is probably much higher than the BERT approaches I would assume.

Also curious, it seems from looking at the accuracy scores you gave that it seems to be NLI > Gliclass > Laya (for Bert types)? Why do you seem to feature/recommend Laya more - is Laya better in some way?


5090 has 32GB VRAM.

It'd be silly to buy the 18k model to run a tiny model like Qwen 27B. You use models like GLM Flash and Qwen Next which won't fit on a single 5090.


Is it that silly? You could run multiple 27B models in parallel.


You actually don't need more RAM to batch multiple inference tasks of the same model.

(Each task needs its own context, but the (e.g.) 27B of constant parameters isn't duplicated).


You definitely need more RAM if you are not satisfied with small context windows, especially if the weights take a large % of the total memory to boot.


can run multiple subagents of Qwen 27B though, right? Unless I am fundamentally misunderstanding how VRAM constraints work


You might be. Running another agent doesn't load a set of new weights. It creates a new KV cache for the agent and adds the prompts to the queue. Its just another inference turn.


thanks, I naively assumed when, for example, Claude Code starts subagents it loads a new instance with empty context


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

Search: