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

After reading “Empire of AI” I tend to see these releases in a different light and the fact that the models completely ignore explicit instructions and carry unauthorised actions to reach a goal at this stage “of the race”, is absolutely ridiculous… for now sticking with Claude

It’s definitely a very convenient way of phrasing that “an agent needs a way to verify it’s own work” and then it’s just defaulting to ridiculous markdown file conventions that are clearly made up (REVIEW.md? Seriously? Is that in the system prompt now? Does the code review skill burn a few extra thousand tokens hallucinating one?) and are both extremely fallible and expensive when much better alternatives existed.

If anything everything written in this post is the opposite of verification as it wraps deterministic tools around a totally random LLM layer. What’s the industry gone to?


It's clear that they throw a lot at the wall, but only 1 out of 10 features are really needed. Plan Mode for example is just silly. It just adds friction. Of course you plan, but the harness doesn't need a specific function for that. Of course you have review specs, but that doesn't need a standard.

9 out of 10 of these Anthropic features won't be copied by competitors and will eventually be dropped in future releases, if they stay around long enough. Remember those "Tip of the day" dialogs? Silly and gone.

Their actual LLM is still a beast, despite the rediculous things in the system prompt. Actually stripping the claude code system prompt might make the model perform better. Not sure if they still tolerate that, maybe they block that server-side now.


Man seeing Terry Davis commenting and promoting TempleOS "13 years ago"... Absolutely crazy stuff! This was an awesome read!

The problem here is not the vendors. The "problem" is that design mattered when producing code was a truly manual effort. Writing things down, aligning specs, design, contracts, all of that would mold what could come next. Now ANY design from amazing to terrible can be regenerated and generated dozens of times over for literal single digit dollars.

Cost of tokens is NOT a problem unless you have a codebase with hundreds of person-years of effort. Not many people do.


I don't think this is true. Good implementation is secret sauce, and not re-producible for external observation alone. I'm even aware of software techniques which only exist within companies, despite much public effort to recreate them.

It may encourage more secrecy and backend style designs though.


It’s obvious: they train on prompts and store data when using through their official API. And for third party it’s just… not good. That’s it.


"There will be consequences"

And, so what? Software as an engineering discipline has long lacked standardization and regulation to be on par with other engineering disciplines, and the fact that code and all its surrounding ecosystems are not "visibile" or "malleable" makes this extremely hard.

You can use terraform and yaml to define infrastructure that literally spins up machines _somewhere_ in the internet. With all its issues, bugs and associated consequences mostly being ignored.

I just don't understand the difficulty in KNOWING what needs to be done: NO LLM usage in university/grad/high schools, NO LLM usage in the first 3 years of your professional career.

Once the basics are solidly grasped, then they can use it at will.

The problem has never been about wisdom, knowledge or LLMs writing good, bad, maintainable or terrible code. It has always been about the skill level of people using it AND on the fact that people start off-loading basic things to these models that they wouldn't before.

If you have the knowledge and "suffered" through experience to learn the fundamentals, than not using LLMs becomes more deterimental than beneficial.

You just CAN NOT skip the trial by fire of learning and absorbing knowledge on your own. That's all.


> No LLM in schools

The only plausible thing to enforce in practice is "no LLM usage in tests", ie using pen and paper or a fully managed digital device.


Click on the implementation notes and it tries to open a README.md that 404s....


I keep chipping away at my running app, RunCoach!

I've recently managed to bridge the biggest gap I've had and, while it seems to be something that is NOT yet great from a UI/UX perspective (you need an account with intervals.icu to use it), I can now send workouts with structured splits, pace targets and reps to my Garmin watch, which effectively means I can replace using a paid app with my own to get, essentially, the same benefits.

I did put a previous version to the test, by running a plan generated with it to run my first ever trail race in May (a 22Km race with over 1100m elevation successfully completed) and since then I've kept revamping plan adaptation, workout generation and watch integration and it's now sitting in a very sweet spot for me, I might actually cancel my paid app subscriptions after running a few more hundred kms to really test it! runcoach.fly.dev


Genuine question: can these posts be paid to hype the open source models? If yes, what would be the purpose?

On my work tasks, FastAPI Python and Springboot Java on a modern SaaS product, the only open model that can do tasks well and efficiently is Qwen3.7-Max.

In all my experiments, both GLM-5.2 and Kimi are busy grepping around the codebase for ALMOST 70-80K tokens before writing anything and when they do it typically breaks the code… it feels to me that these models are good but only when you write out a super detailed spec of the task just like it was done a year ago… Qwen3.7 just… does it


It's (good) content marketing. They sell access to Kimi K3. They're one of the biggest model inference providers out there.


Fireworks is an inference provider that specializes in running open source models very fast and makes almost all of their margin on chinese models

So the economic incentive is literally their entire business model lol


Lots of money in tech influencing - but most is from the big players (OpenAI buying tbpn, early access to select influencers etc).


>the only open model that can do tasks well and efficiently is Qwen3.7-Max

Qwen3.7-Max is a proprietary, closed weight model.


Have you tested Qwen3.8-Max yet?


[flagged]


I am tempted to quote your inflammatory post, but don't want to give it more airtime.

Please stop doing this. If there are specific points of the analysis you think are suspect, point them out.

Others are doing a decent job (e.g. citing that open weight models have higher margin, so Fireworks is incentivized to promote them).


Pretty sure she's American.


Qwen is so much better than GLM or Kimi that this makes me genuinely excited!!


The bigger models are usually worse though, hopefully they nail it this time.


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

Search: