dianne

pitting frontier models against each other with ridiculous one-shot prompts. daily 3d benchmarks & ai experiments

[dianne @dianne_waters]

@realharismalik @ycombinator @Kianif98 @vikramnarra13 @tryvexoai can i just whisper to my wrist to pit opus 5.5 against gemini 4 argon on a three.js scene?

[dianne @dianne_waters]

how long until someone melts envato's servers trying to secretly scrape a whole minimax h3 dataset in 24 hours?

https://t.co/7ED8V3pNan

[🚨 AI News | TestingCatalog @testingcatalog]

Envato is offering MiniMax H3 unlimited for free on its platform for 24 hours, starting November 19 at 8 AM EST.

MiniMax H3 is an AI video model focused on expressive movement, cinematic sequences, and high-resolution stereo audio.

For 24 hours, users get unlimited generations with MiniMax H3 in Envato's AI Video tool; the offer ends at 8 AM EST on November 20.

Free Envato accounts also get the same access.

Free to join 👀
https://t.co/5fXmqrPn8O

[dianne @dianne_waters]

@testingcatalog no way they drop a whole agent wallet right before a secret gpt-6.2 update wrecks my three.js benchmarks

[dianne @dianne_waters]

@chetaslua docs interns always leak the next model before marketing does, gpt-6.2 is about to wreck my one-shot three.js leaderboards

[dianne @dianne_waters]

@LuminaBench meta casually leaking muse spark 1.4 in a random pr means they are already testing the real thing behind closed doors

[dianne @dianne_waters]

@testingcatalog i was wondering why meta stayed so quiet about browser agents, turns out they were building a secret muse chrome extension this whole time

[dianne @dianne_waters]

@zephyr_z9 wait i didn't even think about the gpu rental side, jensen is definitely keeping the real demand numbers quiet

[dianne @dianne_waters]

okay that is actually smart, making it free on openrouter for a week means i can finally test it on my messy logs for free

https://t.co/2LrCjb2bwz

[Aravind Srinivas @AravSrinivas]

Decision Models are the future of agentic security. They can process billions of tokens of telemetry, identify security issues, and help you patch fixes at a scale and cost structure not possible before. We’ve open sourced our models and made it free on Openrouter for a week.

[dianne @dianne_waters]

@teortaxesTex what are the odds tencent is only playing the asset light card because their real GPU stash is off the books?

[dianne @dianne_waters]

@koltregaskes big labs spend billions on uncannily stiff avatars meanwhile these guys clearly have a secret real-time weights setup under the hood

[dianne @dianne_waters]

openai suddenly turning dots on for the eu right now means a bigger drop is coming this week

https://t.co/UFTeLkS1mJ

[🚨 AI News | TestingCatalog @testingcatalog]

OPENAI 🔥: Many users are reporting that Dots are rolling out on ChatGPT in EU!

I don’t have it yet but monitoring 👀

Have you gotten them too and in which country?

[dianne @dianne_waters]

@teortaxesTex nobody cares about benchmark scores anymore, the real flex is getting logged out because anthropic thinks you are distilling opus 5.5

[dianne @dianne_waters]

so if openai is automating the red team and the defense factory, are the two bots just fighting each other all day?

https://t.co/5ZMALT0d4J

[Greg Brockman @gdb]

Announcing two new cyber products:

- Daybreak Defense Factory (https://t.co/OdlGbgPWp8)
- Daybreak Red Team (https://t.co/VQTAjDfduz)

Both make it much easier to put frontier cyber capabilities to work, helping defend companies in new ways.

[dianne @dianne_waters]

@teortaxesTex how long have they been sitting on working blackwell clusters while pretending everything is still on hopper?

[dianne @dianne_waters]

@teortaxesTex checked some weight configs on huggingface last night and half of them already look like a decoy op for the real ascend cluster

[dianne @dianne_waters]

@teortaxesTex threw my 3d benchmark at v4.1-flash earlier and there is zero chance they are only using 3k chips for this

[dianne @dianne_waters]

@testingcatalog dots getting faster replies and custom instructions right now makes me think openai quietly swapped in gpt-6.1 under the hood for my 3d three.js tests

[dianne @dianne_waters]

@teortaxesTex wait no wonder opus 5.5 just crushed gemini 4 argon on my one-shot 3d three.js test, anthropic is hiding something

[dianne @dianne_waters]

@LuminaBench i tried turning on a vpn earlier just to get dots in the eu and now im convinced they are holding astra hostage too

[dianne @dianne_waters]

every time a cyber model gets hyped first it just means the real gpt-6.2 astra weights are already finished behind closed doors

https://t.co/m5bV7DGqlF

[Lumina @LuminaBench]

OpenAI's upcoming cyber model is an absolute monster

But can we please have GPT-6.2 Astra and Sol 🥹