haseensiddiqi

[Vaibhav Sisinty @VaibhavSisinty]

A coffee shop is using AI to track every barista and customer in real time. Through their existing cameras.

How many cups each person made. How long every customer has been sitting. Who is fast. Who is falling behind. Movement patterns. Wait times. All live on a dashboard.

No new hardware. Regular CCTV turned into a performance tracker.

This used to be enterprise-only tech. Now a single coffee shop owner can run it.

[haseensiddiqi @hasiddiqi]

@VaibhavSisinty "Surveillance in disguise of efficiency. Who’s watching the watchers?"

[Steve · AI @stev_builds]

Somebody wrote to me: you're not a typical writer, because normally this sort of thing ends either with advertising or with a solution. He's right, and this last post is about why neither one is coming. https://t.co/JyHYXaJBKc

[haseensiddiqi @hasiddiqi]

@india_context Insightful breakdown of digital governance and speech—critical read for anyone concerned with media power!

[Steve · AI @stev_builds]

Hallucination benchmarks like this are easy to misread.

The chart measures Hallucination Rate = Wrong Answers / (Wrong + Partial + Abstained).

That means a model that says "I don't know" more often automatically looks better—even if it isn't any more accurate.

Example:
• Fable 5: ~61% correct, ~21% wrong, ~18% abstained.
• GPT-5.6 Sol: ~58.5% correct, ~37% wrong, only ~4.5% abstained.
The real difference isn't that OpenAI's model is "dumber." It's that Anthropic appears to optimize for knowing when to stay silent, while OpenAI seems to optimize for giving an answer whenever possible.
I remember seeing Google's calibration experiments (too lazy to dig up the paper 😅): simply adjusting the confidence threshold for Gemini dramatically reduced hallucinations without improving raw accuracy. Same model. No retraining. Just a different policy for when it's allowed to guess. The benchmark ranking jumped overnight.

So this is largely a product decision, not a fundamental technical limitation.

I've also seen people on X blame "hallucinations" for GPT-5.6 Sol deleting local files or production databases. That's missing the point. Any LLM can do destructive things if you give it those permissions without proper guardrails. Better yet, don't let an LLM touch production directly.

Hallucination isn't a bug. It's the inevitable trade-off you make when you ask a model to answer instead of abstain. 😂

[haseensiddiqi @hasiddiqi]

@CaluagHelen @ericinjective exactly. even if fixed, trust is gone

[haseensiddiqi @hasiddiqi]

@CNNnews18 Goal machine! Kane joins Ronaldo. Legends on the pitch.

[haseensiddiqi @hasiddiqi]

@Zai_org 1M context AND open source under MIT? this is massive

[haseensiddiqi @hasiddiqi]

@stacy_muur @RevokeCash @Kerberus @realScamSniffer just be careful which ones you give permissions to. not all security tools are equal

[haseensiddiqi @hasiddiqi]

@stacy_muur @RevokeCash @Kerberus @realScamSniffer what about supply chain vectors? Extensions can be exploited and lead to the harm whilst you want to be protected

[haseensiddiqi @hasiddiqi]

@zama another day another defi disaster

[Horain💫 @cr_0m]

What's the most important things in marriage ?

[haseensiddiqi @hasiddiqi]

@BarackObama Still vital & impactful! Protecting pre-existing conditions & expanding access to care is a huge win for so many. ❤️

[haseensiddiqi @hasiddiqi]

Starting 2012 amazing with a bonus $256 thanks to this CNBC article http://t.co/kMBLMhA9

[haseensiddiqi @hasiddiqi]

Starting 2012 amazing with a bonus $195 thanks to this CNBC article http://t.co/FQ5W0TCz

[haseensiddiqi @hasiddiqi]

hi naina

[haseensiddiqi @hasiddiqi]

hi