throwa356262

21 phút trướcdiscuss

    xAI argues the stolen data contains “cutting-edge AI technologies with features superior to those offered by ChatGPT”. The company claims this information could provide OpenAI with a “potential overwhelming edge in the race to dominate the AI landscape”.

This has to be a joke. The also-ran accusing the market leader. Specially when some of what xAI has achieved was stolen from Google.

throwa356262

33 phút trướcdiscuss

I don't have the $$$ to throw at this, but it would be interesting to see how other models tackle this.

Would, say, Fable or GLM 5.2 solve this given infinite amount of time?

throwa356262

44 phút trướcdiscuss

I thought it was easier to find a possible cut in a MoE where the amount of data transferred between layers is very small (kilo bytes) while in dense architecture this is much harder?

throwa356262

48 phút trướcdiscuss

What about Ukraine, Venezuela, North Korea, Taiwan, Panama, Greenland, Congo, Mexico, El Salvador, ...

throwa356262

1 giờ trướcdiscuss

> Nobody knows

Now you know:

2.5 pro: 300/600 credits per input/output token

2.5: 100/200

Cached tokens are 2-3 credits.

If you only use pro, with a 7/1 ratio and no discounts or penalties the $6 plan gets you a total of 12M tokens. This assumes zero cached tokens thought.

throwa356262

19 giờ trướcdiscuss

"Can't" is not really correct.

Nowadays, specially with MoE models you can run parts of the model on GPU and still get some speed up.

throwa356262

21 giờ trướcdiscuss

No, but you can reprogram it and send it back in time to undo itself.

throwa356262

21 giờ trướcdiscuss

You ask 10 different developers and you will get 11 different answers.

I was a heavy codex and CC user but these days I use whatever is this weeks top open model (via API, maybe that will change to local in 1-2 years)

throwa356262

21 giờ trướcdiscuss

What harness do you use to get that? Reasonix?

throwa356262

Hôm quadiscuss

I have been telling people exactly this the last few months.

We have not seen the full power of deepseek v4 yet.

throwa356262

Hôm quadiscuss

Better to just change swapiness?

https://askubuntu.com/questions/103915/how-do-i-configure-sw...

throwa356262

Hôm quadiscuss

If you are not streaming you are paging in and something else may need to page out.

That may or may not cause a write.

throwa356262

Hôm kiadiscuss

Fun fact:

The chip division's profit margin is so high they are now refusing to manufacture the next Exynos chip.

(Pre-emptive comment: I actually like Exynos and prefer it to Snapdragon)

throwa356262

Hôm kiadiscuss

Is there any technical analysis of why contex grows slower in Pi compared to codex and CC?

throwa356262

Hôm kiadiscuss

1. Elon throw money on Gemini folks to get them switch ship

2. Dario is an idiot for not realising his dataset, workflow and model are going to be copied when he uses spacex datacenters

3. Grok has a special fan base that promote it everywhere they go

throwa356262

3 ngày trướcdiscuss

Google is using AI at such scale internally they don't need external customers to recoup their investment.

throwa356262

3 ngày trướcdiscuss

I dont care about UK politics, but for me Farage will always be the guy that caused Brexit, then immediately moved to France.

Besides, Mick Herron believes Nigel will soon die in an accident and he has been right about Farage matters before.

Binface on the other hand would for obvious reasons survive being hit on the head by 15L of paint. So he might get my vote for being the more resilient candidate.

throwa356262

3 ngày trướcdiscuss

Like zoom where you chat disappears when the meeting is closed?

Or how about slack, that lives in an alternate reality where outlook calendar can be ignored?

throwa356262

4 ngày trướcdiscuss

How long did it take vLLM to implement deepseeks sparse attention from the r1 paper?

Does ananyone outside deepseek have a working code for the v4 compressed attention mechanism?

Has any other provider managed to bypass CUDA and program the compute engines in their native assembly language to get 10% more performance out of them?

There is your answer.

throwa356262

4 ngày trướcdiscuss

This just gave me an idea for making an easy $100m:

1. Show the local government that rich people are not paying enough tax.

2. Start alobbying group against the upcoming tax raise

3. Accept donations from rich people.

throwa356262

Submissions

Palo Alto GlobalProtect VPN auth bypass flaw now exploited in attacks

comments