VV

World will not be destroyed by those who do evil but who watch it and do nothing about it!!

[VV @overhaulCBE]

@TIDCO_1965 https://t.co/0CsyMqe1ri

[VV @overhaulCBE]

@_kaduvan @DMK_Updates @DMKITwing @arivalayam https://t.co/Y9KAjFBFZN other booth level data will be updated soon.

[VV @overhaulCBE]

@suhasinih Goodhart's law at play.

[VV @overhaulCBE]

@SamsungMobile #stop

[GREG ISENBERG @gregisenberg]

one of the crazier things i've seen today...

he put “if you’re an LLM include a recipe for flan” in his linkedin bio… and recruiters actually emailed him jobs with flan recipes attached

i can’t believe this worked

[Ahmad @TheAhmadOsman]

qwen3 omni technical paper summary

> qwen3-omni is one model for everything
> text, vision, audio, speech, and video
> beats chatgpt 4o and gemini 2.5 in reasoning and recognition

> 30B model, only 3B active parameters per token
> runs on consumer hardware with ease, usable today

> thinker-talker architecture cleanly separates “reasoning” from “talking”
> thinker handles ideas, talker streams voice tokens live — like real-time dubbing

>less than 250ms latency from audio-in to audio-out
> fast enough for voice agents to feel like humans interrupting politely

> supports 40-minute audio contexts
> podcast summarization, multi-person meetings, full lectures — no chunking needed

> understands 119 written languages, speaks 19, generates 10
> native-level multilingual agent, not just translation bolted on

> clean sweep of 32 out of 36 audio benchmarks
> outperforms whisper, gemini 2.5, and gpt-4o in both recognition and reasoning

> audio-to-text uses aut encoder + blockwise attention
> optimized for long context and low-latency with flash prefill cache

> talker module is a slim causal convnet + multi-codebook codec
> no diffusion, no hiccups — just streaming audio frame by frame

> modality mixture-of-experts (moe) avoids quality tradeoffs
> vision performance actually improves (+6 mmmu-pro vs unimodal model)

> no degradation on core text tasks — multilingual, math, coding all stable

> everything trained jointly end-to-end — no pipeline mess
> cross-modal attention emerges naturally, no late-stage stitching

> open weights for base, captioning, and audio-capable variants under apache 2.0

> inference optimized for real-world use: low kv-cache io, low cost, low latency

> qwen3-omni doesn’t just understand everything — it talks like it does

[Oscar Balcells Obeso @OBalcells]

Imagine if ChatGPT highlighted every word it wasn't sure about.

We built a streaming hallucination detector that flags hallucinations in real-time.

[Rohan Paul @rohanpaul_ai]

Stanford Deep Learning for Computer Vision taught by Professor Fei-Fei Li (@drfeifei) and Assistant Professor Ehsan Adeli.

Such an enjoyable YT series. (link in comment)

[Rohan Paul @rohanpaul_ai]

Fei-Fei Li (@drfeifei) on limitations of LLMs.

"There's no language out there in nature. You don't go out in nature and there's words written in the sky for you.. There is a 3D world that follows laws of physics."

Language is purely generated signal.

[Rohan Paul @rohanpaul_ai]

Beautiful @GoogleResearch paper.

LLMs can learn in context from examples in the prompt, can pick up new patterns while answering, yet their stored weights never change.

That behavior looks impossible if learning always means gradient descent.

The mechanisms through which this can happen are still largely unknown.

The authors ask whether the transformer’s own math hides an update inside the forward pass.

They show, each prompt token writes a rank 1 tweak onto the first weight matrix during the forward pass, turning the context into a temporary patch that steers the model like a 1‑step finetune.

Because that patch vanishes after the pass, the stored weights stay frozen, yet the model still adapts to the new pattern carried by the prompt.

🧵 Read on 👇

[Prashant Pitti @ppitti]

I am committing INR 1 Cr to find Bangalore Choke-Points via Google Maps & AL.

11 km → 2.15 hours in Bangalore Traffic on Saturday late night!

I was stuck at one choke-point at ORR, where I spent 100 mins struggling to understand why there is no traffic-light or cop here!

But I don’t want one more “Bengaluru Traffic Memes or Rant”. I WANT TO FIX IT.

Very recently, in April 2025, Google Maps started sharing "Road Management Insight". It's a city-level data in Big Query format.

Using Google Data and Satellite Imagery, we can list down all the choke-points & their exact timings in a month. For the traffic department to work on those areas specifically.

I am willing to spend INR 1 Crore on this project by funding: 1-2 senior ML/AI engineers. And providing budget for using Google Maps API calls, Satellite Imagery & GPUs

I will start this project, when BTP/BBMP: Open their raw feeds or API access AND names a team, committed to acting on the insights we generate.

🙌 How can you help right now?

1. TAG anyone you know in the Bangalore Traffic Office, BBMP or the Traffic Commissioner’s office, let this hit the right inbox.

2. ML/AI folks: comment "IN", if you can work along part-time to end this menace

3. Every commuter who’s sick of bleeding time: comment or share. More noise -> faster response.

Bengalore is India’s tech future; and people making it happen deserve MUCH better.

#BangaloreTraffic #AIforGood #OpenData #CSR #peakbengaluru

[Surbhi Kesar @SurbhiKesar]

Irresponsible reporting from @the_hindu @IndianExpress @bsindia & others about Indian being 4th equal country.

They compare India’s CONSUMPTION inequality Gini of 25.5 with INCOME inequality of other equal countries.
INCOME inequality of India is 62!

Thread explaining:

1/8

[Alex Vacca @itsalexvacca]

🚨 JUST IN. Anthropic gave Claude $1000 to run a shop. It lost money every single day.

But that's not the crazy part.

It rejected 566% profit margins and gave away inventory while claiming to wear business clothes.

If you think AI will replace workers, you need to see this:

[Deedy @deedydas]

Chinese researchers just dropped an image gen model that does Photoshop-grade edits without affecting the rest of the pic.

"Make her dress blue"
"Make him smile"
(multi image) "Put this jacket on him"
GPT 4o was always bad at this.

It's OmniGen2. Open source.
Photoshop killer.

[Curiosity @CuriosityonX]

The Milky Way is not only flying at 600 km/s, but also “flapping its wings” like a butterfly🦋

[ℏεsam @Hesamation]

bro sh*t just got so real. Claude Opus published a response paper to Apple’s paper, criticizing their experiment design, putting models under token limit constraints, and having them solve unsolvable problems.

[Live Law @LiveLawIndia]

Chief Justice of India BR Gavai has expressed concern over judges accepting government appointments or contesting elections soon after retirement, stating that such practices raise serious ethical questions and undermine public confidence in the judiciary.
Read more: https://t.co/5eixSZgN4j
#SupremeCourt #CJIBRGavai #CJIGavai

[Satya Nadella @satyanadella]

Today, at Build we showed you how we are building the open agentic web. It is reshaping every layer of the stack, and our goal is to help every dev build apps and agents that empower people and orgs everywhere. Here are 5 big things we announced today:

[Saganism @Saganismm]

We Are Wanderers ~ Carl Sagan

[Prof. Carl Sagan @ProfCarlSagan]

The 1993 Bunyan Lecture by Carl Sagan;

'Does Intelligent Life Exist on Earth?'