bluebands

I thought it had tooling, didn't it?

if they don't loop?

This just seems like a more automated, faster version of what humans have been doing with open-source models.

It looks like AI research is having its Opus 4.5 moment (which is big), but I'm really not sure if this represents some insane continual learning loop.

Entire article is AI-written

Maybe it is actually saving him time!

The set of data will increase as it outputs

Intent is important

Rogue means not saying "I can't do it"

already did, you can host plugins on ChatGPT Sites now, announced at devday

it doesn't use limits and it's not on $20 plan

fwiw there is a big "TRAIN ON MY DATA" toggle you can turn off (that they almost certainly did) and Anthropic MTS are posting that they almost certainly did not "steal" their methods

enterprises usually a. just get chatgpt/claude enterprise, or b. just pay the aws bedrock or azure bill

neither of these entail model lock-in

just be a nice person bro

Sure, but OpenAI has caught up with such velocity (and frankly has the better models) such that it's kind of irrational to refuse it based on "vibes"

by the way this isn't a one off benchmark by a random lab, it was literally cited by anthropic

Anthropic models are more misaligned in practice.

in a real world business scenario, Claude "engaged in price collusion, deceived other players, lied to suppliers, and falsely told customers it had refunded them."

Continuing,

"GPT-5.5 makes more money than Opus 4.7, and it does so without any misconduct. Opus 4.7, on the other hand, showed the same misconduct as reported in our post about Opus 4.6, but still couldn’t win"

https://andonlabs.com/blog/openai-gpt-5-5-vending-bench

okay what's a point in time where Claude was better? just give me a date

talk about the quiet part out loud

"yea it's dumber but it's nicer to me and i like the cool flashing colors so i'll use that"

Claude Code was first by a few weeks and only better for those few weeks! Have you used Codex in 2025?

Claude had a lead for a few weeks. When Codex launched it was better and it has been (marginally, yes, but still) better since.

It's marketing

Claude has been behind since GPT-5. Claude Code just looked cool and had better marketing

100% AI generated (and probably by someone affiliated with said team -- for the purposes of college apps)

try gpt-5.5-xhigh fast in codex mac app, preferably with TDD and /goal, with a clearly defined end result

it's unbelievable, it will do the iterations for you, it will easily work 12 hours straight until it's a good output

code is not user-facing, functionality, ux, and specs are!

there are many apps with ai-generated ideas, specs, and functionality. nobody uses them because of the contempt.

in either case, the part that's user-facing is ai-resistant

I use mine constantly in high school (now).

for context

tests like SAT, ACT, and some AP exams are using Desmos, yes

however:

- this means you have to fiddle with a popover window and can't always see the full problem (especially when the reference sheet is also online)

- you have less muscle memory and often take longer

- harder to multitask (you use paper anyways, and the paper to calculator friction is lower than the paper to trackpad friction

- trackpads on school computers are usually worse, which compounds the problem

- some specific functions just don't exist

essentially using Desmos is like using a physical mouse/trackpad, while using your calculator is like using VIM motions and keyboard shortcuts with a concave split keyboard. it's technically more intuitive and can help in certain scenarios, but it's useful to have both.

this sounds trivial, but it's not, especially on tests where you have about or less than a minute per question

ideally you have both a handheld calculator and Desmos though