nl

49 ms·

841.
by nl 10mo ago
Because the LLMs have only got this good 3 months ago, and market dynamics mean they can't hold them in house without their competitors getting ahead.
842.
by nl 10mo ago
> meant that person would do it in a clandestine fashion rather than this be an agreed upon task prior? Is this how you operate? This doesn't mean this at all In an AI heavy project it's not unusual to have many speculative
843.
by nl 10mo ago
I don't understand what your argument is. It seems to be "people keep saying the models are good"? That's true. They are. And the reason people keep saying it is because the frontier of what they do keeps getting pushed
844.
by nl 10mo ago
> OpenAI's sales and marketing expenses increased to _$2 billion_ in the first half of 2025. I believe they include the costs of free ChatGPT user's in that $2B. Worth it considering the conversion rate they are getting (5-6% i
845.
by nl 10mo ago
He's a pretty decent programmer. It's interesting that some months ago when his nanochat project came out the HN Anti-AI crowd celebrated him saying "I tried to use claude/codex agents a few times but they just didn&#x27
846.
by nl 10mo ago
Don't believe the what you read about dropping birthdates somehow being a disaster. The US has always been an immigrant nation and despite the current administration that is still how the military is manned.
847.
by nl 10mo ago
Yes! The Swordfish - completely right. Not sure why I remembered otherwise. Was the Swordfish successful? I think it was used at Narvik but it was slow and not very manoeuvrable.
848.
by nl 10mo ago
Worth noting that the attack on the Bismarck was by biplane Gloucester gladiators which were outdated even at the start of the war. Compare them to the planes that carried out attacks in the Pacific theatre. The Grumman Avenger was maybe 2
849.
by nl 10mo ago
People are cheaper than all that tech. Unmanned drones make sense because they are more capable. That's not the case with most ships.
850.
by nl 10mo ago
This will work fine for bash commands, but most Agent implementations also have read/write file functions that are implemented using local file operations.
851.
by nl 10mo ago
I think their point is more that that architecture of this CodeRunner program isn't very clear. It's unclear if it is a container manager, or comes with a LLM Agent built in. These are two separate concerns and the README makes it
852.
by nl 10mo ago
Windows 2 was an interesting moment in computing history. Microsoft wasn't the dominant player and was sort of the underdog in many ways. Lotus was usually considered a more important company (Lotus 123 was huge) and WordStar dominated
853.
by nl 10mo ago
There's lots of interesting tooling in my space. Currently open in my browser are https://github.com/liquidmetal-dev/flintlock https://github.com/e2b-dev https://www.daytona.io https:
854.
by nl 10mo ago
This runs on GCP The issue with using raw VMs is you want fast startup. If you are running hundreds of pieces of code per hour as you develop, or have 10 or 20 agents running simultaneously it's much better to have something faster to
855.
by nl 10mo ago
I'm after this too. I want to have a "container" (used in the conceptual sense here - I'm aware of the differences between container and other solutions) that I can let an AI agent run commands in but is safely sandboxed
856.
by nl 10mo ago
I don't see how that follows. uv doesn't exactly stop python package supply chain attacks...
857.
by nl 10mo ago
> Groq exclusive license non-exclusive license actually.
858.
by nl 10mo ago
Give him The Kingkiller Chronicle to finish first please!
859.
by nl 10mo ago
The reason they are called "world models" is because the internal representation of what they display represents a "world" instead of a video frame or image. The model needs to "understand" geometry and physics
860.
by nl 10mo ago
> handicap Your words. Because it's very good tech for inference? It doesn't even do training. And most inference providers for Open Source models use NVIDIA eg Fireworks, Basten, TogetherAI etc. Most NVIDIA sales go to trainin
861.
by nl 10mo ago
The founder created TPUs at Google. > Prior to founding Groq, Jonathan began what became Google’s Tensor Processing Unit (TPU) as a 20% project where he designed and implemented the core elements of the first generation TPU chip.
862.
by nl 10mo ago
Given that Groq is basically the TPU spin out Google should have done years ago it shows what a valuable asset TPUs are in Google. Still this they should spin that out though!
863.
by nl 10mo ago
GPT 5 Codex is great - the best coding model around except maybe for Opus. I'd like more speed but prefer more quality than more speed.
864.
by nl 10mo ago
NVIDIA release some of the best open source models around. Almost all open source models are trained and mostly run on NVIDIA hardware. Open source is great for NVIDIA. They want more open source, not less. Commoditize your complement is bu
865.
by nl 10mo ago
Of course! But the operational costs are much lower than some people in this thread seem to think. You can find a safe margin for the price by looking at aggregators. https://gpus.io/gpus/h100 is showing $1.83/hou
866.
by nl 10mo ago
> which is funded by the dumping What dumping do you mean? Are you implying NVidia is selling H200s below cost? If not then you might be interested to see that Deepseek has released there inference costs here: https://github.c
867.
by nl 10mo ago
The ongoing costs are mostly power, and aren't that massive compared to the investment. No one is leaving an H100 cluster not running because the power costs too much - this is why remnants markets like Vast.ai exist.
868.
by nl 10mo ago
Lots of sources, and you can do the math yourself. Here's a few good ones: https://github.com/deepseek-ai/open-infra-index/blob/main/20... (suggests Deepseek is making 80% raw margin on inference)
869.
by nl 10mo ago
I'm not longer CTO at an AI startup. Updated, but don't actually see how that is relevant. > All going into the hands of a small group of people that will soon need to pay the piper. It's not very small! On the inference s
870.
by nl 10mo ago
I don't see why that follows. The “Rs in strawberry problem” is literally "count the token R" in the word "strawberry". One could argue that the learnt tokenization model where it is tokenized into 3 tokens (see

More ›