Sigmaboy King

[Sigmaboy King @VIVEX2q]

If you pasted API keys, secrets, private repo tokens, customer data, or anything sensitive into Ox Alpha (especially through agents), treat those as potentially compromised. Rotate the keys. Don’t assume the “no training” promise protected you.

[Sigmaboy King @VIVEX2q]

If you pasted API keys, secrets, private repo tokens, customer data, or anything sensitive into Ox Alpha (especially through agents), treat those as potentially compromised. Rotate the keys. Don’t assume the “no training” promise protected you.

[Sigmaboy King @VIVEX2q]

The Ox Alpha “zero data retention / we don’t train on your data” claim were never clean.OpenRouter’s own listing said prompts & completions were retained by the provider. OpenCode said zero retention. The Stealth terms had conflicting language that opened the door to collection

[Sigmaboy King @VIVEX2q]

They’re not a random startup. They have the scale, the incentives, and the data. Whether they train on it or not is something only they can verify — and we can’t.

[Sigmaboy King @VIVEX2q]

They’re not a random startup. They have the scale, the incentives, and the data. Whether they train on it or not is something only they can verify — and we can’t.

[Sigmaboy King @VIVEX2q]

Now that https://t.co/yzZpM8dTAy (Zhipu) has revealed it was them all along, that data is in their hands. A Chinese lab ran a massive free anonymous stress test, collected real production traffic (including a ton of coding-agent sessions), and is now saying “trust us"

[Sigmaboy King @VIVEX2q]

The Ox Alpha “zero data retention / we don’t train on your data” claim were never clean.OpenRouter’s own listing said prompts & completions were retained by the provider. OpenCode said zero retention. The Stealth terms had conflicting language that opened the door to collection

[Sigmaboy King @VIVEX2q]

@bridgemindai I think they should launch Opus 5.1 instead of Fable 5.1.

They need to build more hype for Fable and not launch it every single week, because if they launch it this week or next week, it will probably only be a 10% improvement. They should build more hype for Fable.

[Sigmaboy King @VIVEX2q]

GLM 5.3 flash is not even that good, its only good on benchmarks. The glm team loves BENCHMAXXING

[Sam Altman @sama]

i think we should do another party for our next model release, the 5.5 party was a lot of fun.

what would make the next one awesome?

[Sigmaboy King @VIVEX2q]

@shiri_shh USA GDP IS 40 TRILLION$

[Sigmaboy King @VIVEX2q]

Claude is the apple of AI

[Sigmaboy King @VIVEX2q]

@0x0SojalSec ts does not even beat opus 4.8😭😭

[Sigmaboy King @VIVEX2q]

@trikcode Yea, its the best on 95% benchmarks, I feel like people just want views or standout.

[Sigmaboy King @VIVEX2q]

@jp54362 @grok Do people actually hate Anthropic?

[Sigmaboy King @VIVEX2q]

@jp54362 No one hates andropic btw. Stop spreading your agenda

[Sigmaboy King @VIVEX2q]

@AI_IA_liker @sattyyouneed are you aware that its more servere if your using chinese models yk. I'm fine with giving America my data but Not china.

[Sigmaboy King @VIVEX2q]

@Jan_Sakir @sattyyouneed bro, we can NOT be trusting china with our info

[Alexander Panfilov @kotekjedi_ml]

We can finally talk about it:

We found a way to extract hidden reasoning of frontier models using a vulnerability in the APIs of every frontier AI company.

We verified that our reasoning token count matches billed API thinking tokens 1:1 for most of the prompts we queried.