dr_kiszonka

Contributors: GCSS had the original idea. JPP tried to talk him out of it. JPP did the first literature search but GCSS lost it. GCSS drafted the manuscript but JPP deleted all the best jokes. GCSS is the guarantor, and JPP says it serves him right.

But the chat version does change over time, correct? It has been my experience that Sol's performance has deteriorated significantly.

Very curious that everyone here (so far) seems to assume this dashboard presents real data.

It looks fun! Reminds me a bit of TLDRAW computer.

Same here. 5.5 would sometimes oneshot relatively complex models even without MCP. Opus, however, even with Fusion 360 MCP access would usually make comically bad models. It seemed that Opus tried to reason through the problem while GPT would come up with a general plan of using whatever tools were available to solve the problem. In chat, it would use the available Python and trimesh. It was also the only time I had ChatGPT work on something for over 30 min without giving up.

Thanks!

Thank you for your work on Snap! and for finding the time to respond to our comments.

What device did you get your nephew? When I was looking for one, those listed on the Make Code website were pretty bad and expensive (or unavailable). I would love to find an affordable retro handheld like Miyoo or Anbernic with Make Code support.

This is much better than I anticipated. I don't mind AI that much, but this version of HN surfaces really interesting articles I've missed.

So I know a Lisp dialect? Fascinating.

The vocals in the few songs I listened to are so very clearly AI. I find Suno's much more credible in many genres.

If anyone has tried Suno's Studio 2, I would love to hear your take on it. At $24-30/month, it's approaching Ableton Live 12 Suite rent-to-own territory, without the "own" part.

Could anyone explain to me the difference between thinking traces ("intermediate tokens") and the final responses? Specifically, why is it that Claude Opus 5's reasoning in Code is very easy to follow and sounds quite natural, while its answers are full of these very annoying AI-isms and sentence fragments that are void of meaning?

Are thinking traces and final answers trained for different objectives?

I spotted Claude doing it when accessing PubMed.

We also have sanctions and incentives to induce specific behaviors, but they don't apply to agents. We can put a muzzle on Guile 3.8 but we can't turn it into Genuine 3..8.

Nice! And it continues the lineage of https://en.wikipedia.org/wiki/Zachary%27s_karate_club

old.reddit now requires a login :/

I gave up on Antigravity when Google changed its rules around paid plans AI quotas, leading to my AI Pro getting depleted in up to four rounds.

Have they ever gotten around to fixing it or is Antigravity only feasible with their Ultra plan?

"Well, actually," it is usually not just HTML, but I am with you on liking these single-file apps. Security is complicated because these pages often rely on importing online packages, which can get poisoned in the blink of an eye.

Nvidia Reportedly Weighs Perplexity Investment at More Than $30 Billion Valuation

https://finance.yahoo.com/technology/ai/articles/nvidia-repo...

I got their Pro subscription in some promotion. Is it just me, or is Perplexity generally very bad? Specific models are often unavailable, and context collapses after only a few rounds. It frequently routes me to their Computer, which requires buying credits (upselling). The Pro search is very fast but typically very superficial.

Does anyone have a good use case for Perplexity?

>Also, sometimes older models work fine, Opus 4.8/4.6/4.5 are worth trying.

I would go as far as saying they are sometimes better than the newer ones! It is impressive that Opus 5 can one-shot a complex website, but it is very annoying if I only asked it to write tests.

Last week, I asked Sonnet 5 to add one line to the documentation, commit, and push. Ten minutes later, I discovered it embarked on a side quest of exploring my whole Git history, different branches, etc.

Interesting product. Are you really using Fable 5 to generate sounds? FYI, in Chrome for Android, the only way to hide those bottom sheets is to tap the handle. Grabbing and sliding them down, reloads the page, which is a little annoying.

Why are we regurgitating the past? How did it even make it to the home page?

(And why is this the top comment?!)

I find video models very hard to steer. They can output amazing "shiny" things quickly, but if you want something very specific, you may need to spend days on it. It is not like sitting down with an artist and seeing how they make the exact changes you need in real time while also ensuring artistic cohesion.

AI is obviously very useful, but perhaps not in the way people think. Two aspects I would like to see improved are instruction following and providing relevant feedback.

> the point of learning should not be to do something, but to expand your own mind and understanding of the world (and beyond)

I agree with your comment overall, but people have different reasons for learning and it may not be productive to tell them they are learning for the wrong reason. Let's celebrate all learning for I fear we are heading in a direction where it will become increasingly uncommon.

"An increased risk of Guillain-Barré syndrome (severe muscle weakness) was observed after vaccination with SHINGRIX"

https://www.shingrix.com/side-effects/

I get that people see this and think: ads and social media. My first thought was cognitive neurorehabilitation and brain stimulation.

Realistically, probably ads, but maybe not only that?

(AI start-up idea: one of our ads a day keeps dementia away! /s)

Cooperative Thread Array

Only if you have a wide enough screen. I had to rotate my phone to landscape. Thanks for the pointer!

I agree with you. But do you remember when we could chat with a colleague about how much of a doofus our boss was without worrying that some automated system would possibly notify the boss about it? Nowadays big brother is always watching.