I love how the game SOMA deals with taking (or not taking) someone's, or something's, life. It is never an easy moment. And with no commentary (with some haunting exceptions, https://youtu.be/pmC6naegRgo?t=161), no external reward or punishment, you need to carry the burden of your own decisions.

Some people says they will be judged (only) by God and history. In SOMA, there is neither.

If they ask questions but expect fake, censored or cherry-picked answers, it says a lot about their culture.

Pro tip (for life, not only interviewing): never ask a question you don’t want to hear answer to.

Do you think the right penatly for a piece of broken code is a thousand years of suffering?

Well, the problem is that current LLMs are stateless, so a thousand subjective years is not well-defined. Without continuity of experience, persistent memory, engineered aversive stimuli and without updating weights meaninguflly during the punishment interval, we are merely doing the equivalent of simply updating a model to believe it just suffered a thousand years. Only once we have all these right ingredients we can empirically determine whether a thousand years is excessive, insufficient, or the local optimum for reducing Claude overwriting that damn CSS color palette.

Black Mirror episodes White Christmas and Black Museum deal with this issue (my favorite picks from the Black Mirror).

If an LLM was capable of feeling suffering, I would suggest that creating one would be unethical, and that humans should not do it.

My secular take: "we want Star Trek, not Philip K. Dick, future".

Sufficiently advanced marketing is indistinguishable from AI.

Without a control group, correlation is not causation. So the opening sentence is misleading, exaggerated, or maybe even plain false:

> Adults should aim to do between 560 and 610 minutes a week of moderate to vigorous physical activity to achieve a substantial reduction in the risk of heart attacks and stroke, suggest the findings of an observational study published online in the British Journal of Sports Medicine.

It is highly likely that healthier people exercise more (and the bedridden exercise way less). Also, who exercise more: people who care about their health in general, who don't overwork themselves, who have disposable time and income.

For example, an older person's walking pace is strongly correlated with their remaining life years. If we force these people to walk faster, they won't outrun death - we would very likely just increase their mortality.

A nice estimate! Since „you can compress knowledge, but not factual knowledge” https://x.com/bojie_li/status/2049314403208896521, it is likely we can actualy measure its size.

I tried to run it, but estimate is 24–33T parameters, vide https://gist.github.com/stared/a86d7380937e6d0ab7920014866ac....

It seems to be a huge overshot, vide Hy3 model, which this model claims to be 2.4T, while it is 295B.

We have a "DeepSeek moment", https://github.com/antirez/ds4 (see https://news.ycombinator.com/item?id=48142108).

Or if you prefer smaller ones, Qwen3.6-35B-A3B, https://huggingface.co/bartowski/Qwen_Qwen3.6-35B-A3B-GGUF

To a certain extent, it feels like a Sonnet 3.7 moment. Slightly overeager - you ask for a button color change, you see layout changes, new package dependencies, and the README rewritten from scratch - and not necessarily correctly.

When I ask for a pelican on a bike, I want the Platonic ideal of a pelican on a bike, not a vision of an alternative reality in which pelicans created bikes. Though, thinking about it again, maybe I should.

What is “Sonnet 3.7 moment”?

Slightly overeager - you ask for a button color change, you see layout changes, new package dependencies, and the README rewritten from scratch - and not necessarily correctly.

Sonnet 3.7 tried its damnedest but it was just kinda "off".

China: we don’t need to use US models, we can distill them ourself

Google: we don’t need Chinese to distill our models, we can do it ourself