Naman

880 posts

Naman

Naman

@make_it_so

Guy

New Delhi, India Bergabung Mayıs 2010
602 Mengikuti113 Pengikut
Tweet Disematkan
Naman
Naman@make_it_so·
@DiscussingFilm Fwiw I had built this exact feature to support prime video. My service supports Netflix and Disney+ Hotstar as well. Always iterating. It would be great if you guys could check it out. watchflix.group
English
0
0
0
0
Naman
Naman@make_it_so·
@capajj @thdxr This has been my experience as well. It will “actually actually actually” its way around a simple problem
English
0
0
5
256
Capaj
Capaj@capajj·
@thdxr It's also the only LLM to fail this simple question
Capaj tweet media
Sokolnice, Česká republika 🇨🇿 English
4
0
14
2.3K
dax
dax@thdxr·
deepseek is outrageously cheap even if you ignore the current 75% discount input token prices are 35x cheaper than opus cached tokens are 178x cheaper
English
183
157
4.8K
207.8K
Naman
Naman@make_it_so·
@tankots Hindi ad in blr. 🍿
Filipino
0
0
0
26
Tanay Kothari
Tanay Kothari@tankots·
we just discovered a hidden marketing opportunity no one is paying attention to. and it's happening in the streets of bangalore. US tech companies are burning millions on digital ads to reach bangalore. but the entire city is sitting in traffic, looking at the back of an auto rickshaw. not their phones. bangalore has some of the worst traffic in the world. the average person spends hours a day inside it. and almost no one in tech is using it to their advantage. so for @wisprflow's launch in india, we did something different. this week, 100 autos are driving through bangalore wrapped in wispr flow branding. if you're in the city this week, there's a chance the auto you're stuck behind is one of ours. send us a photo when you spot it!
Tanay Kothari tweet mediaTanay Kothari tweet mediaTanay Kothari tweet media
English
429
225
2.8K
736.7K
Naman
Naman@make_it_so·
@rishabhm @FahreedBalsara Yep and justice system is the US is so good right? Thats why you can’t even abort a rape related pregnancy in half the states
English
0
0
0
30
Rishabh Mukherjee
Rishabh Mukherjee@rishabhm·
@FahreedBalsara Judges in the US are the exact opposite of this. They are pretty exlicit about their political leanings. Yet nobody is lining up to ask judges to recuse themselves because they are federalists or libertarians or whatever. This is just plain stupid stuff from Kejriwal.
English
4
0
4
178
Naman
Naman@make_it_so·
@Parikshit_K_ Good whey is the same price as this pack and has the same amount of protein. Moreover it is important to have different sources of protein in your diet
English
0
0
0
224
Schindler Rao Shinde
Schindler Rao Shinde@Parikshit_K_·
Instead of spending on 11 other things on high protein items, just have one scoop extra whey. You shouldn't depend on labels to fulfil quota Most of the high protein stuff you see is just marketing & justification to sell it at premium.
SidheshPathre|Health Coach@PathreSidhesh

It’s good that many Brands are coming up with High Protein options. In the next 5 years i am positive that we can have the tools necessary to solve protein deficiency in India. Awareness still a challenge i feel.

English
12
3
120
10K
Naman
Naman@make_it_so·
@nahcrof either way appreciate you jumping on this. The performance of the models was grea before the issues. I was hitting >100 tps!
English
0
0
0
9
Naman
Naman@make_it_so·
@nahcrof i had this issue happen for k2.6 as well but only once. but then again I only got the sub today and was using deepseek mostly and gave up after running into “No content” issues once
English
1
0
0
83
nahcrof
nahcrof@nahcrof·
Alright, I've been aware of the weird issues with the new deployment and they should mostly be fixed - deepseek-v4-pro, and the precision models should not "explode" and just give up responding for the rest of the session anymore - issues with hermes that were resulting in empty responses should be fixed as well (there is an issue relating to infra where occasionally a response will get cut short, but it seems minimal and it's one of those things that might take a deep dive to fix, I'll try to have that improved by tomorrow) Thank you to everyone who sticks around even when I end up allowing bugs like this to slip through
English
18
0
83
3K
Naman
Naman@make_it_so·
@ShaukeenBharat @chiragbarjatya Check this one. It says “made with milk” on the pack. This exact rule probably helped push them to not scam probably. They’ve made these ice creams for a while but their reputation will probably stick
English
0
0
0
66
Naman
Naman@make_it_so·
@nahcrof seeing a lot of "No content" responses from this model. Seems to go away on clearing context. Using it with the opencode integration
English
1
0
0
229
nahcrof
nahcrof@nahcrof·
deepseek-v4-pro is up on CrofAI!! $1.15/m input $0.23/m cache $3.00/m output and we'll see how this goes but it only costs 1 request for those using a subscription (if I have to switch it over to an only precision model I'll post here ahead of time)
English
26
6
175
9.5K
Naman
Naman@make_it_so·
Nobody is safe from AI. the shit I saw makes me want to pull my eyes out.
English
0
0
0
3
Vikram
Vikram@vchennai2·
I am no longer solo Introducing our incredible CTO @evanvolgas - Data Eng for 20 years @netflix, @Mailchimp - Shipped 32 PRs his first day - One of the best people I know We're building the best database sandboxes in the world. Clone any postgres DB in <6s And we will win
Vikram@vchennai2

I got into @ycombinator solo After 7 rejections Before the batch: -$2.1M raised alone -Sold 200K ARR -6M+ views I'm building the first database sandbox @ArdentAI We let you infinitely clone any Postgres DB in <6s so coding agents can test code on a 1:1 of prod Time to win

English
41
5
306
29.2K
Naman
Naman@make_it_so·
@ZackKorman Maybe a oauth desktop client secret accidentally added to the agents context window?
English
0
0
0
479
Zack Korman
Zack Korman@ZackKorman·
What a long winded-way to say the employee had a passwords.txt file in their Google Drive.
Zack Korman tweet media
English
17
22
525
44.5K
Naman
Naman@make_it_so·
@a6mhr @markmeyourze Ive been to Nari and Kage, love that place. I wanna try the award winning cheese
English
0
0
0
117
Ami Palan
Ami Palan@markmeyourze·
I’ve been consuming Eleftheria cheese for about six years now. During Covid, I used to get it delivered from their Powai unit to my house via Porter almost twice a week. I’m a huge fan of their burrata, the quality and the taste are truly next level. But what makes Eleftheria even more special is the people behind it. They’re not just passionate about what they do; you can genuinely feel the care, craftsmanship, and dedication that goes into every product they make.
Narendra Modi@narendramodi

Cheese from India makes its mark globally… India made an impressive debut at the Mundial do Queijo do Brasil 2026, which is a vibrant international competition for cheese and dairy products. Four Indian products won medals, including 1 Super Gold, 2 Golds and 1 Silver. The Super Gold was won by Eleftheria Gulmarg (Brie Style), the Golds were won by Yak Churpi-Soft, Nordic Farm, Leh, Ladakh and Eleftheria Brunost (Whey Cheese) while the Silver was won by Eleftheria Kaali Miri (Belper Knolle Style). Congratulations to Mausam Narang and Thenlay Nurboo. Such successes strengthen India’s artisanal dairy sector on the world stage.

English
10
13
250
41.3K
Naman
Naman@make_it_so·
@ku1deep ok this might be a dumb question but you don't overcommit gpus on k8s (if not time slicing). so you assign one gpu to one job. how is this a concern? and in what circumstances would you time slice a gpu meant to do model training?
English
0
0
0
4
kuldeep
kuldeep@ku1deep·
4/ The most important IMO is k8s/Slurm layer. it is the boring middle bit. It maps a CUDA context to a pod to a Slurm job to a team. Sounds dull. But it's what turns "GPU 3 is hot" into "Anita's fine-tuning job on GPU 3 is hot and it's cost us $847 so far today." That's the bridge.
English
2
0
4
283
kuldeep
kuldeep@ku1deep·
You asked why so here goes : 1/ GPU observability is weirdly hard. You get utilization graphs from one tool, NCCL logs from another, inference metrics from a third, and nothing tells you which team's job is burning the H100 at 3am. Not Fun. So we built l9gpu. 🧵
kuldeep@ku1deep

So the team @last9io just shipped a GPU monitoring stack and like all things Last9 it is performant and can deliver at least 4 nines of monitoring reliability. AT LEAST. Now we built it with a few design partners and new stuff us getting added fast but I am looking for more design partner. Do you have a GPU monitoring workload we can help. You won't pay us anything and you will have my gratitude.

English
3
7
51
8.5K
Naman
Naman@make_it_so·
@nityasnotes I am this dude but I don’t work at exa unfortunately
English
0
0
1
1.2K
Nitya Sridhar
Nitya Sridhar@nityasnotes·
there is ~1 infra dude at exa that finds dev tools he likes then optimises the entire experience of using it internally, adds permissions, keeps schilling it till everyone tries it this dude thinking your thing is good is probably worth hundreds of thousands of dollars
English
20
6
756
50.4K
Naman
Naman@make_it_so·
@nexxeln Banoffee if youre still looking
English
0
0
1
318
nexxel
nexxel@nexxeln·
im in blr where should i get brunch? preferably in or near kormangala
English
22
0
94
12.3K
Naman
Naman@make_it_so·
@thdxr Yeah i like hex. But then you dictate more than 5 words and opencode shows as “pasted x lines” with no way to see if it got it right. pls fix
English
0
0
0
59
dax
dax@thdxr·
the biggest impact on my coding workflow lately hasn't been anything agent related it's realizing how good these parakeet local voice models are and then just dictating everything into opencode typing is more of a burden/bottleneck than you realize
English
151
36
1.6K
100.7K