Kia Khezeli

[Kia Khezeli @k_khezeli]

@rh_fox @deweytn Great question. Probably not, post processing with LTX or H3

[Kia Khezeli @k_khezeli]

@robrombach Great work! Noticed that the model can generate thermal images. Have you benchmarked its radiometric fidelity?

[Kia Khezeli @k_khezeli]

@SebastienBubeck Build GPT-5.7

[Kia Khezeli @k_khezeli]

@PhilSalesses link: https://t.co/d5BlE5wbbh

[Kia Khezeli @k_khezeli]

@PhilSalesses That's a reasonable guess considering the recent Google case.

[Kia Khezeli @k_khezeli]

We are entering the fifth day of complete blackout of communications in Iran. But as George Washington said "Truth will ultimately prevail where pains is taken to bring it to light.”
#DigitalBlackoutIran

[NetBlocks @netblocks]

⚠️ Update: #Iran has now been offline for 96 hours, limiting reporting and accountability over civilian deaths as Iranians protest and demand change; fixed-line internet, mobile data and calls are disabled, while other communication means are also increasingly being targeted ⌛️

[Kia Khezeli @k_khezeli]

@SebastienBubeck GPT-5.2 Pro for research feels so magical I wish I were still a PhD student.

[Kia Khezeli @k_khezeli]

@alec_helbling Nice work. Do you have any intuition for how this extends to small objects or background concepts?

For example, in your clip the saliency map for “tree” doesn’t cover the tiny background trees. Is the ConceptAttention signal inherently biased toward the most prominent instance?

[Kia Khezeli @k_khezeli]

Imagine creating a computer vision model by just describing it. That future feels closer than ever.

[YRIKKA @YRIKKA_Inc]

New research from the Yrikka team: Text-to-Model (T2M).

You can now create a custom object detection models with nothing more than a prompt and a few raw images.

No datasets. No labels. No annotation grind.

[Kia Khezeli @k_khezeli]

@Yuchenj_UW Yes, but 3 out of the 4 authors on the original GPT paper did have PhDs: Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. OpenAI in 2018 operated much more like an academic research lab than a tech company.

Same for FAIR when they released PyTorch.

[Kia Khezeli @k_khezeli]

@PascalUnger LFG

[Kia Khezeli @k_khezeli]

@amasad It is a transient interface that if adopted at scale provides data to OpenAI to improve text-to-agent reliability

[Kia Khezeli @k_khezeli]

@fchollet “With four parameters I can fit an elephant, and with five I can make him wiggle his trunk.” John von Neumann

[Kia Khezeli @k_khezeli]

Every week, a new AI-slop newsletter that I never subscribed to hits my inbox. It feels like the “I had pizza for lunch” phase of early social media.

Trivial content is just not engaging. Here’s to moving on.

[Kia Khezeli @k_khezeli]

@grok CausalARC is an extension of the ARC-1 benchmark with a fully specified causal world model.

https://t.co/j0bYvcdXzJ

[Kia Khezeli @k_khezeli]

In CausalARC, Grok-4 is the only model that sustains high accuracy on logical reasoning tasks: Discovering causal rules and scaling them to larger arrays.

This robustness likely reflects how it was trained, with stronger inductive biases for causal/world-model reasoning.

[Kia Khezeli @k_khezeli]

@ycombinator @EasyCodeAI Congrats on the launch @PaulTheLi

[Kia Khezeli @k_khezeli]

Congrats to the @krea_ai team for their incredible work improving the aesthetics and realism of FLUX. @MaxClouser shows Krea's frequency signature is almost identical to real images, much closer than the FLUX Dev.

[Max Clouser @MaxClouser]

FLUX.1 Krea is out! I compared it to FLUX.1 Dev in the frequency domain using 50 real NYC photos and found Krea aligns closer to reality. Check it out:
https://t.co/eDRpX3hen4

[Kia Khezeli @k_khezeli]

@itsPaulAi Is this open source or open weights?

[Kia Khezeli @k_khezeli]

@fchollet I learned some basic Spanish a few years ago but gradually got less out of it as they started over-optimizing for user engagement and not enough for learning.