polz
33.3K posts
@ClaudeDevs Framing this as a free credit instead of a regression for users is wild.
Now I have to make the Claude Code experience on T3 Code significantly worse in order to keep using the 25x subsidization provided on a Claude Code subscription.
x.com/theo/status/20…
Theo - t3.gg@theo
Solution incoming…
English
Starting June 15, paid Claude plans can claim a dedicated monthly credit for programmatic usage.
The credit covers usage of:
- Claude Agent SDK
- claude -p
- Claude Code GitHub Actions
- Third-party apps built on the Agent SDK
English
@0xUnihax0r Someone said the other day, if I remember reading correctly,
That there’s a 0day of some sort on ERC20 floating around lol
Assuming it must have some specific conditions ofc,
Was it an old aged wallet?
~ Dr. Axius.
English
@flowersslop ever since sonnet 3.7 every speculation of a new model is "agi" lol
English
I am now 90% sure GPT-5.5 in some openai native openclaw like harness, which seems to be the plan, will be AGI by most definitions, certainly by mine.
It will be called 5.5, not 6, but I am not alone with the opinion that it should be callled GPT-6, but whatever, it will be AGI
English
@ExodusPgame @justalexoki Yes, but it could be compacted into a tool call when time is relevant, like how Claude only researches when it’s relevant
English
@justalexoki Anything you add to the LLM context makes it constantly want to comment on that.
Gemini has knowledge on time and it comments on it all the time and it's annoying.
English
@sporadica back then, if there was a model with this much of a jump, there was always change, sonnet 3.7 became the first model people even trusted in the ide and it didnt even have this big of a jump
English
@sporadica im mixed, on one hand, it feels and looks like any other ai hype when a model drops
but i think there is some merit to the hype because of how big the increment is, recently its only been a few % bumps with models. this is different
English
really interesting "boy who cried wolf" thing happening in the AI space rn
yea, a lot of earlier releases (GPT-3, o1, o3, Opus, etc) were met with insane fanfare (and crazy dooming, thanks yudkowksy and co) and so far have changed little.
i myself have been responsible for some of that insane fanfare, too, i know it
but now people are so tired of the hype they will instinctually gut react that any major step change in capabilities is really a nothing-burger. the AI hype-ists have cried wolf too many times apparently.
i don't really know what to make of this, but i'll be honest that to me this time feels different. maybe i'm wrong. time will tell.
can@marmaduke091
The duality of man
English
Soon, every piece of software in the world will have their vulnerabilities exposed. And then shortly after, no software will have vulnerabilities.
Theo - t3.gg@theo
I would highly, highly recommend you make sure your phone, computer, browser, and important apps are all updated and on the latest versions.
English
I was expecting Mythos in May and nextGPT by April 14th. Little bit surprised they got beaten to the punch
Tibo@thsottiaux
@adonis_singh Uhm
English
FYI, mythos/opus can one shot in a day what I used to take a week to write an exploit. You can probably pwn most electron apps using opus now!
s1r1us (mohan)@S1r1u5_
Tip, pick any electron app out there and write an exploit fo it and you will learn a lot. Stuff I learned while doing this - v8 inline cache - a bit of v8 new heap sandbox implementation - a bit of v8 regexp bytecode interpreter - ROP chain in MacOS ARM64(kinda complicated than x86) - ROP chain in Windows x64 - using windbg, lldb, radare2, ROPgadget
English
@linusoidal i feel like this has been said every model drop since like sonnet 3.7 lol
English
@scaling01 i agree that there obv wont be a open source 10t model, but tbf, if there is 1 thing china is good at, its benchmaxxing
so expect similar scores but not similar performance
English
I will take the opposite side:
- there won't be an open-source 10T model
- and there also won't be an open-source model that has equal or better scores on all benchmarks in < 12 months
Mckay Wrigley@mckaywrigley
society needs to grapple with the reality of a mythos-level model being open source in <12 months. i’m not sure we are prepared.
English
@polzerise Fair, hopefully well get to mess around with it soon through cloud.google.com/blog/products/…
English
I feel like this graph is some what nonsense tbh, AI with proper tools and time can write a lot of exploits I feel rn. The only way this makes sense is if they didn't give opus nearly enough time or the right tools and mythos far more?
also anthropic claims "every single one was confirmed to be a true positive" during their Firefox testing.
Honest to god has anyone seen similar performance? Cuz from my experience its not even close to this, its finds bugs without a doubt but it def doesn't isn't void of false positives.
English
Idk if they made other changes, but I almost exclusively use Claude Code for fixing things on my system and now it's refusing to do that? 🙃
English
@NoahKingJr imagine how it felt for the microsoft employees that were assigned to develop it lol
English
I believe education should be free and accessible. Hardly paid anything during my journey in my career as a Pentester, nor have I asked for money to share knowledge.
You could learn any tech skill for free, tbh.
random guy on the internet@KharayKrayKray
“I just bought a developer course for…” My guy, if you want to get scammed so bad, just go trade forex.
English