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

the author seems unaware of the SAT/SMT solver/analysis ecosystem

"Please don't post shallow dismissals, especially of other people's work. A good critical comment teaches us something."

https://news.ycombinator.com/newsguidelines.html


Why do you think so? I didn't get such impression.

No. Did you even read the article? It talks about the "specification gap", which is the difference between the formalized semantics and the intended semantics.

Every formal method has that problem (including the mentioned trivial ones like SAT and SMT).


Opus 4.7 told me an open source program had a bug, but when i asked it for help crafting a PR or toy implementation it refused and told me i was violating Claudes TOS. I tried to plead for it to give only the most innocuous example that could not possibly work except by illustration but it continued to refuse. it would only discuss, not write any single piece of related code.

Using Claude Max was fun for more than a year but the last weeks i'm constantly fighting their Harness, their weekly TOS changes and outages. Anthropic lost all goodwill with me as a developer. I'm switching to OAI.

only if we crypto first

good god, lemon


Anyway, why does Microsoft bottleneck itself when it could have 10 different AI teams. That's why 10 new AI startups can achieve what these behemoths can't.


Microsoft already suffers from lack of software consolidation. It costs a lot of money to maintain 10 different tech stacks. Not to mention having separate security/privacy review processes for each.


We are talking multitrillion company here, money shouldn't be an issue. Though you prove my point, they lack coordination. Also AI lacks coordination and strategy and won't solve BigTech management problems any time soon. Best they can try is providing tools trying to capture what others, that are actually inventive, can do. That's why they will stay at Azure level.


The demo didn't finish loading on a 4090 after 5 minutes...


Which demo did you try? The car configurator? Also if you're using Firefox, WebGPU isn't supported there yet, so I recommend trying using a Chromium browser.


DevTools will most likely help with debugging.

In my case, several of the WebGPU demos on https://simplystream.com/demos fail (while keeping on falsely showing "Loading") with the error message (in DevTools):

"WebGPU requestDevice failed, trying low spec DOMException: Failed to execute 'requestDevice' on 'GPUAdapter': Required limit (64) is greater than the supported limit (32)."


Was using latest Chrome


what you'd really need to do is actually a custom --style that is trained on her images specifically, which midjourney allows

https://www.youtube.com/watch?v=Iy4d4UP7w2c


this is what I get explicitly asking for "Cat and Girl" in the style of Dorothy Gambrell.

https://cdn.discordapp.com/attachments/1098302916395270235/1...


It gets even worse being MORE explicit about referencing THE '"Cat and Girl" online comic strip by Dorothy Gambrell'


More or less. Cat and Girl is interpreted as 2 not-proper nouns. But even when you just spam Dorothy Gambrell it doesn't get it. It doesn't seem to know the style individually, at all.


That’s what I was afraid of.

This comic reads like someone complaining about an imaginary problem, not a real one.

(The images generated here are lovely though)


"Ssssssssss," the basilisk said, with "seeming" approval.


You don't have a way to reach out on your profile but I'd be willing to scan and send you a csv of your collection for easy selling for a low flat rate


This is my test. Just got 5gbe fiber installed this week.

https://imgur.com/a/fTG815R


Consider applying for YC's Summer 2026 batch! Applications are open till May 4

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

Search: