Nirant @NirantK

trying my best to escape the permanent underclass with friends nirantk.com localhost Joined December 2012

Tweets

13K
Followers

17K
Following

2K
Likes

49K

Nirant @NirantK

30 minutes ago

@pHequals7 @anmolm_ is this your family?

0 0 0 41 0

View Details

Nirant @NirantK

34 minutes ago

@noheeriye @Somya_Crazy levels to this wealth thing huh

0 0 0 14 0

View Details

Nirant @NirantK

11 hours ago

I've been wrecked by bitter lesson every year of my career except 2023-2026 when I saw it coming 😂

people forget why the bitter lesson is actually bitter. it is because the person confronted by it likely thought the bitter lesson did not apply to them. otherwise it is quite a cheerful lesson actually!

0 0 12 1K 2

1 0 10 1K 1

View Details

Nirant @NirantK

2 days ago

@shubhamratrey @heykavana worth keeping notifs on for you!

0 0 1 75 0

View Details

Shubh @shubhamratrey

2 days ago

1,166,400,000,000 (1.17 trillion) tokens. Every month. Ignition. @heykavana 📈🚀

1 2 6 250 0

View Details

Nirant @NirantK

2 days ago

I expect that 9/10 companies making CUA apps are RL env fodder for the labs and they won't capture any customer $$ 1y out

1 0 11 2K 1

View Details

Nirant @NirantK

2 days ago

@sasank51 Would love to dropby soon man!

1 0 3 175 0

View Details

Nirant @NirantK

2 days ago

@dbreunig @SrihariSriraman Great to see this Srihari!

0 0 2 358 0

View Details

Nirant @NirantK

2 days ago

@skeptrune in what way

0 0 1 28 0

View Details

Nirant @NirantK

2 days ago

@dillon_mulroy @davekiss ohh god, thanks for this, p50 anxiety went down

0 0 3 7K 0

View Details

Nirant @NirantK

2 days ago

@arkobuilds Given how buggy Dia is, you're right to feel that way

0 0 0 41 0

View Details

Keshav Jindal @Keshavatearth

3 days ago

If you don't want to hire a team and want your employees to mostly do what they do best, you might be interested in this. I'll only be able to work with 3-4 teams (10-30 people teams) for this to be high quality.

Keshav Jindal @Keshavatearth

3 days ago

I kinda agree with Dan. For big companies, yes they need a team but if you're less than a 50 people, you want just one-person (+ founder conviction) who doesn't even need to be on your full time payroll. And yes, this is a pitch for myself -- before you hire someone, talk to me.

0 2 13 1K 6

0 3 24 7K 29

View Details

Nirant @NirantK

2 days ago

@i2cjak praying that this is a jokepost secretly

0 0 3 870 0

View Details

Nirant @NirantK

2 days ago

@skeptrune haha, godspeed on this train!

0 0 0 13 0

View Details

Nirant @NirantK

2 days ago

This is clearly a well thought out argument but incomplete. Let me break this down in 2 parts: 1. Open Source 2. Whether we want to pretend regulation can stop an arms race or not Let's talk OSS first: Beyond Sonnet 5.5/Sol 5.6 level of models, of which China will continue to make plenty and has already made few — there is nothing which regulating OSS gets us in terms of security risk surface areas. This is about surface area, better models do change the attack rate because it lowers the compute/$ per attack. Second, What should be the right coordinated/societal response to increased attacks? Is it to build antibiotics and better software (including agents!) which can respond to fast, deep and longer chain attacks? Would you like to pretend that an enemy state like Pakistan be regulated? Or actually staff a modern Armed Force which is trained & equipped? I think the right time to regulate OSS was 2023-25 and that has long past gone. Either way, I don't think OSS or not makes a LARGE difference here. I say this with the understanding that abliterated models increase the surface area AND make it easier to find more exploits. I agree with the case @skeptrune has made elsewhere and Varun has made here. It's just that the solution to abliterated models is not regulating GPUs or whatever because it's an arms race against a mix of anonymous, innocent (think REA of Photoshop) and enemy state actors. And the only way to arms race is to pick up and fight.

Varun Mayya @waiting4_asi

3 weeks ago

I've done a complete U-turn on my opinion on open source AI. We should be careful (and probably disallow release) of models on par with current open source Astra/Opus level models. AI is now reverse engineering games from binaries and remaking them. Reverse engineering games is

614 239 4K 557K 1K

1 1 14 3K 9

View Details

Nirant @NirantK

2 days ago

@fayazara @shobhitic @Cloudflare Hey Fayaz! Heard back on email, would be really really grateful for relief from the Cloudflare for Startups team!

0 0 0 40 0

View Details

Nirant @NirantK

3 days ago

Hey @Cloudflare, can you help here? My card paid $6773 out of the blue? There are no budget or default alerts even! No warning. I was hoping to finish these experiments and then refactor for cost. I am migrating workloads from others to here. I can't afford that sudden bill.

8 5 64 19K 15

View Details

Nirant @NirantK

2 days ago

Great to see video frames support!

Nischay Dhankhar @nischay_twt

2 days ago

What happens when you heavily distil a model with millions of decision based samples? I am super excited to release Drex Max, it's so far the strongest decision model which is top #3 in almost all the public benchmarks we evaluated. It can handle text, images, and video frames.

2 2 23 1K 11

0 0 2 803 0

View Details

Nirant @NirantK

3 days ago

@Keshavatearth chewing an ice cube works for 30-60 min

1 0 0 206 0

View Details

Nirant @NirantK

3 days ago

@NocodeTalks @Railway I mean, there's always AWS.

3 0 3 1K 0

View Details

Nirant @NirantK

3 days ago

I run production systems on @Railway. We found atleast 2 of our keys to be compromised: AWS and Anthropic. Here is what I am doing: 1. All services, including user facing & crons are off. 2. Rotating all our credentials. And asking our customers to rotate them too. Damn embarrassing but has to be done. By Sunday night IST, I will complete the migration to @render + @Cloudflare and have our customers enter new keys.