Amogha retweetledi
I create a web application where you can read random pages from old Kannada magazines, it's really fun and interesting. Give it a try. #kannada
Check here : kannada-archive.netlify.app
English
Amogha
424 posts
@DivaJain2 kaam dham hai nai kuch toh movie hi dekhoge na kolkata mei
bombay is too busy carrying the country
Indonesia
In Calcutta Nandan 1 is full even for the screening of an obscure Soviet film.
Barely a couple of hundred have shown up for a legendary film like "Le Samourai" at Regal with capacity of ~1000.
Bombay for all its wealth will always be a Vada Paav class city :)
Film Heritage Foundation@FHF_Official
A line of cinephiles formed to watch Jean-Pierre Melville's landmark French film 'Le Samouraï' (1967) today at Regal Cinema, Mumbai!!
English
@IndraVahan There's a ton of classical ML work being done in enterprises. It just doesn't make it to twitter
English
so freakin true. where are all the classic ml people? where did they all go?
finbarr@finbarrtimbers
I can’t stop thinking about how 1) RL is the dominant paradigm in AI and everyone is designing around it but 2) the old school RL people are almost all uninterested and are sitting the LLM stage out on the sidelines Not what I would have predicted 10 years ago
English
Not surprising at all! The Deccan was a Kannada speaking region for a very long time.
Amit || ಅಮಿತ್@infinite_shunya
Latur was purely a Kannada speaking region before Maratha kingdom rule. The name itself has ‘ur’ (ಊರ) which is a Kannada word
English
> giants
> ignored
never before since GPT-OSS have I seen so much aura farming for a mid-tier open source model. Just drop the tech report already.
Tokenization for Indian languages works the same as for all other, you just train your tokenizer on the appropriate distribution.
Parimal@Fintech03
Sarvam’s 105B model (and its smaller 2B sibling, Sarvam-1) solved a problem that global giants like OpenAI & DeepSeek ignored: the Tokenization Tax. In most AI models, an Indian word like Namaste takes 4-8 tokens to process, while an English word takes only 1.4 & this makes AI 4x more expensive for us. Sarvam built a custom tokenizer from the ground up that brought Indian language fertility down to 1.4 tokens/word, matching English for the first time in history. This is how 40 people outperformed others in local efficiency :)
English
VC investment in subbammanangadi wen
Shailesh Mahato@8M_Shailesh
🚨 Sweet Karam Coffee raising Rs 30 Cr in Series A1 The company’s valuation is expected to rise by 85% to Rs 580 Cr Revenue 4X to Rs 46 Cr (FY25 In April 2025, the company raised $8 Mn in a Series round led by Peak XV Partners Read more @entrackr #Fund entrackr.com/exclusive/excl…
HT
@SatanicThot666 @NarasimhaKan @sandeepssrin It's restocked once or twice a year. There's an hmt showroom in MG road
English
Today HMT is no more.
Here's my favorite watch of all time - the HMT Gandaberunda manual wound.
The last of the lot of manual wound mechanical watches made in India....and one of the only watches ever made in Kannada.
RIP HMT
thehindubusinessline.com/news/variety/t…
English
@NarasimhaKan @sandeepssrin FR, I've been trying to get it but no luck :/
English
it’s always the performative, coloured hair, tote bag, coloured half-sleeve shirt, “artsy”, “anti-capitalist”, pedostache, bibliophile, cinephile, social justice keyboard warriors who haven’t seen a gym or a real paycheck in their whole life
English
Amogha retweetledi
Been tinkering with DeepSeek OCR. Doesn't seem to work that well on wireless tables. Take a look at this section from an investment report.
I wouldn't call this SOTA like some people on X are calling it, but it's a great model! It produces 400-500 Tok/s on an A100 with vllm and really strong performance overall.
It's also very strong at discriminating between text and figures. The lines get often blurry when figures have a lot of text in them but this model correctly classifies flowcharts/charts with text as figures. This is important for building document ingestion pipelines for research papers.
We haven't benchmarked DeepSeek OCR yet at @tensorlake on our internal datasets which has widely used documents from insurance, healthcare, energy and financial services. Much better than other open source models but I think it's not going to beat Dots.OCR yet.
English
Half of my for you feed is filled with people lamenting the poetic couple tweet
bas karo bhai samajh gaya sabko pyaar nahi milta
English
Amogha retweetledi
The unbridled joy of listening to someone smart who’s not trying to sell you anything.
Dwarkesh Patel@dwarkesh_sp
The @karpathy interview 0:00:00 – AGI is still a decade away 0:30:33 – LLM cognitive deficits 0:40:53 – RL is terrible 0:50:26 – How do humans learn? 1:07:13 – AGI will blend into 2% GDP growth 1:18:24 – ASI 1:33:38 – Evolution of intelligence & culture 1:43:43 - Why self driving took so long 1:57:08 - Future of education Look up Dwarkesh Podcast on YouTube, Apple Podcasts, Spotify, etc. Enjoy!
English