Hugsun

That can only be true if the workload is compute bound, not memory bandwidth bound.

Iceland has mostly been governed by the center right conservative independence party, which has been quick to sell land and resources to anyone interested. Kárahnjúkar is the canonical example of this, a huge hydro plant powering aluminium smelters owned by a foreign company.

Right now, different parties have the majority rule, and their interest in projects like these are not clear. I would suspect that a motivated investor could fairly easily get them built. The hurdles would be logistics and connectivity much more than red tape.

Iceland is culturally and politically scandinavian with some influence from the US. In august there will be a vote to start accession talks with the EU. This has been a heavily contested issue for years, largely due to Iceland's unique resources.

Is there anything of value in this project?

It sounds interesting at a glance, but it seems to be AI slop. So it's hard to tell if there are any interesting discoveries there, or just some worthless results described with performatively advanced language.

Have you tried Qwen 3.6 or Gemma 4? They're not frontier level but certainly have their uses.

Note that your monitor is a QD-OLED, not a QLED monitor.

Despite the similar names, these are two very different technologies. QD-OLED is a OLED panel that uses QD (quantum dots) to achieve a wide color gamut. QLED is a LED backlit LCD panel that uses QD in a similar way.

The difference in performance characteristics of these two panel types is dominated by the difference between OLED and LCD.

That HN agent review is pretty usable. Very cool!

It is unbelievable that this is a controversial opinion.

That seems like a decently sized market. Maybe not for an AI lab though.

Two things can be true at once.

The whole post and that paragraph in particular sound AI generated, that triple "No" is a big tell. I'd not be surprised if that confirmation complaint is just a random suggestion proposed by the AI that wrote this.

They're comparing Qwen's moe vs dense (smaller difference) against Gemma's moe vs dense (bigger difference). Your proposed alternative misses the point.

Unfortunately, llama.cpp quantization technology has been stagnant for two years. The main quantization developer left or was kicked out of llama.cpp due to an attribution dispute. He created his own fork ik_llama.cpp where he has made multiple new and better quants.

unsloth and byteshape are just using and highlighting features that have been available the whole time. I am very invested in figuring out a solution to this dispute, or some way to get the new quants upstreamed.

LLMs produce a distribution of token probabilities which is then sampled. This sampling is the only random part of the system.

If you just take the most probable token every time, the system becomes fully deterministic. We don't do this as the output becomes more stiff and less creative.

SCOTUS is indeed compromised.

I heard from a medium reliable source that this loophole wasn't as popular as the zeitgeist implies. I'd love to know how true that is and if so, how the rich finance themselves.

I listened to a few episodes on a recommendation. The episodes lead you to believe that the host is some agnostic person that's just curious about these reports. They are not as the original lead is from another superstition believer podcast.

They also present the cameraman as a token skeptic, who is of course quickly swayed into belief.

They lean heavily on a host of tricks with long histories of non-reproduction when tested rigorously.

A "scientist" (known crackpot and woo believer) is employed to make the experiments sound. And their terrible academic reputation was explained away using conspiratorial arguments.

I found TT wholly unconvincing and consider it a scam to get people to pay for the actual evidence. I won't pay of course and confidently assume it to be poor based on the publicly available material.

I've found GPT-5 Thinking to perform worse than o3 did in tasks of a similar nature. It makes more bad assumptions that de-rail the train of thought.

The version of llama.cpp that Llamafile uses supports structured outputs. Don't waste your time with bloat like langchain.

Think about why langchain has dozens of adapters that are all targeting services that describe themselves as OAI compatible, Llamafile included.

I'd bet you could point some of them at Llamafile and get structured outputs.

Note that they can be made 100% reliable when done properly. They're not done properly in this article.

I gave up after it didn't let me see the prompt that went into the LLM, without using their proprietary service. I'd recommend just using the API directly. They're very simple. There might be some simpler wrapper library if you want all the providers and can't be bothered to implement the support for each. Vercel's ai-sdk seems decent for JS.

I was under the impression that this is not the case. Aspartame has been studied a lot and not found to be harmful.

I discovered that this is very common when posting a long article about LLM reasoning. Half the comments spoke of the exact things in the article as if they were original ideas.

I was very pleased to discover that Mistral's Le Chat has inbuilt support for python code execution and sympy is importable.

It will regularly use it and reliably when asked to.

Consensus seems to be forming around the fact that Ollama is not genuinely trying to be part of the open-source community.

Ollama are known for their lack of transparency, poor attribution and anti-user decisions.

I was surprised to see the amount of attribution in this post. They've been catching quite a bit of flack for this so they might be adjusting.

I've experienced this a lot as well. I also just yesterday had an interesting argument with claude.

It put an expensive API call inside a useEffect hook. I wanted the call elsewhere and it fought me on it pretty aggressively. Instead of removing the call, it started changing comments and function names to say that the call was just loading already fetched data from a cache (which was not true). I could not find a way to tell it to remove that API call from the useEffect hook, It just wrote more and more motivated excuses in the surrounding comments. It would have been very funny if it weren't so expensive.

I have not yet found any compelling evidence that suggests that there are limits to the maximum intelligence of a next token predictor.

Models can be trained to generate tokens with many different meanings, including visual, auditory, textual, and locomotive. Those alone seem sufficient to emulate a human to me.

It would certainly be cool to integrate some subsystems like a symbolic reasoner or calculator or something, but the bitter lesson tells us that we'd be better off just waiting for advancements in computing power.

How are you using them? Who is enforcing the daily quota?

How long is it's thinking time when compared to o1?

The naming would suggest that o1-pro is just o1 with more time to reason. The API pricing makes that less obvious. Are they charging for the thinking tokens? If so, why is it so much more expensive if there are just more thinking tokens anyways?

Really!? Do you have a source for this? This would be really interesting if true.

What sort of propaganda do you forsee this enabling?