Hacker Newsnew | past | comments | ask | show | jobs | submit | schmorptron's commentslogin

I'm giving them the benefit of the doubt and interpreting it in a charitable way because they sound earnest about it, this is incredibly ambitious and cool-sounding, and I wish them all the best. It's something that's some sort of pipe dream, a noninvasive diagnosis machine that is able to use certain generic measurements and then derive insane levels of data from it. We've of course seen Theranos, but the holy grail remains.

Of course, there's always the tradeoff between research data collection and access vs user privacy, and striking that balance is incredibly hard. To make anything like this even remotely feasible you'll need a shitton of data and have it fully available to your researchers as well, while somehow safeguarding individual users. anonymizing medical data is impossible without rendering it near useless. Hoping they can figure that out! (Also, with human bodies being so different from one another, combatting bias is probably an eternal challenge)


you can build the datacenter right next to the tank and use the now-warm cooling water to pump into the tanks!

Cursor's composer models are finetuned kimi

They are unusable (unless you want to deliberately destroy your codebase). So if Cursor's models are Kimi based, then well. I'll skip them altogether.

Kimi works great in their CLI, but their CLI has a number of workarounds for quirks of their models, including detecting when the model gets into a loop, and reverting to a checkpoint but letting the model compose a "message" to its past self (search their CLI for "BackToTheFuture"...) It doesn't work so well in a harness that doesn't take those quirks into account.

I'm using Composer extensively, and it works great for me. Your experiences are not universal.

They are far from unusable. They aork great for 80-90% of a typical full stack dev. Alot less useful for more noche stuff

I wouldn't skip at least testing the original. Model distilling done by Cursor could be the culprit.

Composer is really good, but just like any Chinese model it needs a good plan. It's cheap and fast, in 1 month of pro I used the equivalent of 500$ in API credit for it.

Composer 1.x was poor. The new one is a totally different beast and absolutely fine for day to day.

They're not unusable, they're just bad when compared with all the real frontier models.

I only use composer 2.5 day to day and it works fine with human review.

It's "just" an opencode fork but it adds some nice features to try out while not being a full orchestrator metapackage like oh-my-opencode. Quite nice! Though it would be even nicer if this stuff came upstream or as an easy extension instead in the future

What would a diffusing reasoning model look like? have a pre-defined length [thinking] block that gets diffused over a long time, and then the final output block uses what is in that thinking block as part of its input? And how do diffusion models decide the output length in the first place, is it a pre-set parameter? or does it diffuse an [end] token into the middle somewhere?

got one answer by reading the rest of the comments, makes sense that the diffusion process is inherently reasoning-like: https://www.inceptionlabs.ai/blog/introducing-mercury-2

The irony of "we train on all of humanity's collective output, but god forbid anyone trains on ours" is still incredible

All these people know is greed. It's in their DNA.

Capitalism is designed to promote greed. It's the central point of our society's current design.

Cool project! I'll be trying it out. I've been a big fan of throwing whatever sources I have on a new topic i'm trying to get into into a llm "project" and then asking it to teach me, grounded on the actual content to speed things up.

But at the same time, I'm afraid getting everything laid out for you in exactly the way you want will erode some of the understanding you build by going through a primary source directly and figuring things out the hard way. So this having more focus on actually doing stuff by yourself seems right up my alley (while still tending to the LLM induced intellecutal laziness... ) .


Maybe this will replace raptor-mini as the "free" model on copilot plans? (but I don't see it at all yet on the student plan, in vscode or the cli)


the new intel ultra whatevername 3 series seems to come a bit closer there, so the framework pro with its explicit linux support might be an option


It's interesting that (for example for the explore agent https://github.com/Piebald-AI/claude-code-system-prompts/blo... ) they use a personality "you are a file search specialist" and "your strengths" framing. I thought that was largely thought to be useless, or even counterproductive nowadays? Does anyone know more about this stuff?


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: