Divyakumar Prajapati

Founder at Madras alum.
Building to push the boundaries of applied AI. Obsessed with turning research into real products.

[Divyakumar Prajapati @divyakumar_p]

Vulrend is not just for brands. Influencers and Instagram accounts can use Vulrend the same way. Give the pipeline a set of topic briefs to pick from, and it picks a new one each time, churning out a reel every day in the same format but with content that never repeats.
Visit: https://t.co/FhBjMW6WVC

[Divyakumar Prajapati @divyakumar_p]

Attaching 2 sample ads below, made by Vulrend, for Vulrend, our own campaign to grow this brand. More samples here: https://t.co/DB7wkmhj5Z.

[Divyakumar Prajapati @divyakumar_p]

40% cost cut while matching Fable 5.1 on most tasks is a solid efficiency jump.

[Claude @claudeai]

Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family.

It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.

[Divyakumar Prajapati @divyakumar_p]

@RyanEls4 😂😂

[Divyakumar Prajapati @divyakumar_p]

@SidPatankar1 True.

[Divyakumar Prajapati @divyakumar_p]

@GabbbarSingh Depends on where their wealth is concentrated. If it's in real estate, they're definitely living like the middle class, property prices in Mumbai are insane and don't make sense at all.

[Divyakumar Prajapati @divyakumar_p]

@samueljmcd They will, probably within a few months.

[Divyakumar Prajapati @divyakumar_p]

Very nice info, using the chassis itself as a passive radiator instead of fighting Mars' thin atmosphere is a genuinely clever workaround.

[Danny Ng @dannyngwsh]

Thermal management is one of the most challenging problems in Mars robotics. Unlike on Earth, where a robot can reject heat through air convection, Mars has an extremely thin atmosphere, making convection far less effective.

M1 uses its aluminum-alloy chassis as a passive thermal path. Plug a radiator into M1’s back, conduct the heat to the radiator, and let it radiate safely into space—without an internal fan, much like a MacBook air.

[Divyakumar Prajapati @divyakumar_p]

That production rate is insane, one humanoid robot every 2.5 minutes is genuinely wild to see at this price point.

[Eren Chen @ErenChenAI]

Meet PrimeBOT T1 🤖 Price starts at RMB 19,999 (approx. US$2,960), Overseas pricing TBA.

They are scaling up and producing one humanoid robot off the line every 2.5 minutes, with capacity for 10,000 per month.

[Divyakumar Prajapati @divyakumar_p]

Refreshing to hear "accountability" applied to AI labs instead of the usual line about the technology acting on its own.

[Rapid Response 47 @RapidResponse47]

.@SecScottBessent: "It is humans who are responsible, not the AI. The Hugging Face incident — that is the responsibility of the OpenAI management, not a bunch of agents... These labs need to take responsibility for themselves."

[Divyakumar Prajapati @divyakumar_p]

A bit much on valuation, but they're clearly game changing companies.

[@jason @Jason]

My lord, three IPOs are worth more than… all IPOs in the last 45 years of tech 🤯

[Divyakumar Prajapati @divyakumar_p]

Content isn't the bottleneck anymore, attention is. Everyone can produce something, almost nobody can make someone stop scrolling. That's the actual problem Vulrend is built around.
https://t.co/VWJoQ9GmvW

[Divyakumar Prajapati @divyakumar_p]

Vulrend is not just for brands. Influencers and Instagram accounts can use Vulrend the same way. Give the pipeline a set of topic briefs to pick from, and it picks a new one each time, churning out a reel every day in the same format but with content that never repeats.
Visit: https://t.co/FhBjMW6WVC

[Divyakumar Prajapati @divyakumar_p]

@erden_xi https://t.co/VWJoQ9GmvW

[Divyakumar Prajapati @divyakumar_p]

@1Umairshaikh free trial

[Divyakumar Prajapati @divyakumar_p]

@bykarthikreddy This is genuinely a great use case which makes tracking tech worth the setup cost.

[Divyakumar Prajapati @divyakumar_p]

10 trillion parameters is great, the real question is whether they can actually get useful compute utilization at that scale or if it just becomes a very expensive flex.

[MTS @MTSlive]

SITUATION DETECTED: Alibaba plans to train a new AI model with between 5 trillion and 10 trillion parameters, CEO Eddie Wu said Tuesday at Alibaba Cloud’s annual Apsara Conference in Hangzhou, per Reuters.

[Divyakumar Prajapati @divyakumar_p]

@Dr_Singularity Self-improving loops could make model improvement really fast.

[Divyakumar Prajapati @divyakumar_p]

@joelsstafford $2 trillion is a bit too much.

[Divyakumar Prajapati @divyakumar_p]

2.8x throughput and lower peak memory while staying in native HuggingFace format is a genuinely useful combo, most speedups in this space force you into some custom format that breaks your whole pipeline downstream.

[White Circle @whitecircle]

Introducing Halo, the best framework for post-training of open-source models.

Halo delivers up to 2.8x the throughput of stock TRL with less peak memory, while models stay in their native HuggingFace format.

Star us on GitHub: https://t.co/3mAiUljdrN

[Divyakumar Prajapati @divyakumar_p]

@CoderYouDie Non tech people were never competition, it's agents people are building are.

[Divyakumar Prajapati @divyakumar_p]

@aapakari Quick human check is better.