Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ

Computer Programmer

[Ludonauta @LudensLudonauta]

I was studying MCPs and this comment really blew my mind.

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

I fell asleep at 4.37AM, still bedtime is 30/30. Wtf is that supposed to mean lol? @Apple

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

why don’t they release performance metrics like the other labs?

if the claims are indeed true, this will be huge.

[Alexander Whedon @alex_whedon]

Introducing SubQ - a major breakthrough in LLM intelligence.

It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA),

And the first frontier model with a 12 million token context window which is:

- 52x faster than FlashAttention at 1MM tokens
- Less than 5% the cost of Opus

Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention).

Only a small fraction actually matter.

@subquadratic finds and focuses only on the ones that do.

That's nearly 1,000x less compute and a new way for LLMs to scale.

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

listening to an ai generated podcast by @NotebookLM and there was not one mention of Gemini, while tons of references to OpenAI and Anthropic. lol.

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

crazy save. crazy driver as well.

sad to see he’s not winning these days :(

[Formula 1 @F1]

WHAT. A. SAVE.

Max Verstappen 🤯

#F1 #MiamiGP

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

@BoazWith That’s the plan. I’d go a step further and restrict network access to a few select domains with a sidecar proxy or something of that sorts.

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

is anyone using autonomous coding agents that are self serving? at an org scale? I know stripe does it and calls them "minions" lol .

i've found a couple of OSS solutions but they are a bit murky security-wise, i will patch them though.

i know Devin has a solution, but what's the fun in using cloud services, i'll raw dog self host the system.

hmu if you have any experience ( good or bad ) with this pls :)

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

@0bytematt Tricky spot to be in!

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

another day, another one of my personal projects killed.

i'm starting to understand that connecting third party products that you don't own to these agents is not going to work as a product in itself. if the market is big enough, it's only a matter of time before the big guns are coming for you.

this loss feels personal, going back to the drawing board

[Claude @claudeai]

Claude now connects to the tools creative professionals already use.

With the new Blender connector, you can debug a scene, build new tools, or batch-apply changes across every object, directly from Claude.

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

AI was supposed to be great at converting unstructured data into structured, actionable decisions. So the question is - are enterprises actually able to leverage AI at their scale, when all their knowledge is sitting inside large, not-so-well-formatted 2000s themed PDF files?

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

agents can't read your company's homework

Every time an enterprise brings in a new vendor, someone has to sit down and explain how everything works. Documents, processes, playbooks. Again. For every single vendor.

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

Even if vendors build this capability, they still need to cross-contextualize across thousands of documents, index them, link them semantically. It's a ton of work and frankly really uninteresting to build, when they can focus on improving their actual product. And if not done right, next month's AI bill eats into their lunch money - especially for small vendors.

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

How are founders today predicting what to build next? How do they know that there is enough demand for what they're planning?

Are there any tools that they use? If yes, then what tools?

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

Tell me he's not @thelonelyisland

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

3b1b is great, but this video of @SebastianLague explaining neural nets is insanely good, and simple.
https://t.co/jYLQFZdDyM

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

Realising that you’re parents may not always be right, but they have your best interests at heart is adulthood imo

[c. @vers_150516]

This might be THE max verstappen edit cause oh my lord I am on knees after seeing this

[Karan Bataviaᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠᅠ @KaranBatavia4]

Read somewhere that Linus Torvalds faced problems with the operating systems he used; and hence developed an entire OS himself.

I'm curious as to how people point out those flaws and actually start working on them, with me I just say "it is what it is" and move on with my day.