melfie

I’m still glad we have Wikipedia as a foundation instead of a publicly traded corporation. The fact that there are way worse offenders doesn’t exempt them from scrutiny, though.

I wouldn’t trust them enough to run anything important.

Qwen 3.6 still works fine on my gaming PC that didn’t cost anywhere near that. ¯\(ツ)/¯

The article mentions the largest producers, so I wonder how much of this is people buying from local breweries instead of drinking Budweiser.

This opinion piece from the BMJ entitled “The illusion of evidence based medicine” sets the stage: https://www.bmj.com/content/376/bmj.o702.

There is so much corruption and industry capture that it can be genuinely difficult to separate the real, unbiased science from “Big Pharma make number go up”. Telling people they are uneducated dipshits and to just “follow the science” ignores and further degrades the fundamental lack of trust.

Eh, I might wait and see how the benchmarks stack up before I sell my house to buy 8 B200s.

On a serious note, what the hell is going on at Alibaba? Now that now that Lin Junyang is gone, 3.6 was the last release we can actually run and now nobody is releasing billion parameter models anymore? Hope the community gets really good at distillation. ☹️

Open weight is analogous to a compiled binary. Similar to how Windows is closed source, but still runs in your own hardware, whereas Linux is truly open source.

So I’m not sure what the marginal cost per launch is. I remember reading that the propellant between Starship and Supeheavy is something like $100M per launch, which would put 20 launches at around $2B and would make the re-usability not actually a cost saver vs. an expendable rocket. However, I just read the article below that says the launch cost is more like $3-$5M, so I may have been completely off base in my original assumption.

https://spaceorbitals.com/articles/starship-v3-cost-per-kg-2026/

“OpenAI” signed the open weight pledge, which only makes sense given the company name and original values. They’ve had gpt-oss for quite a while, but it feels like a pretty half-assed effort for the purposes of virtue signaling. I’m not holding my breath for them to release an open weight model on par with GLM, Kimi, or Qwen.

Yeah, there’s a lot of innovation going on to run powerful models on modest hardware and the state of the art for running local models is changing all the time. If the trend continues, models hosted in a data center will only be for niche use cases.

I thought SpaceX was at least on the right track with Starship, but then I read this article talking about how even one trip to the Moon could take up to 20 launches:

https://www.nextbigfuture.com/2023/12/nasa-says-up-to-20-spacex-starship-refueling-launches-per-moon-mission.html