@TheSkayeth Both will continue to be billionaires even if categorically wrong.
English
RuskEater
327 posts
Something I told 13 yo: Cities inhale and exhale each generation. People move to cities in their 20s in search of colleagues and mates, move back out to raise their kids, and then when their kids are in their 20s, they return.
English
What are the odds that the CEOs of the 3 big AI companies are :
Altman ( Description of AI)
Amodei ( Starts with A, ends with I)
Pichai (Ends with AI)
English
A hypergrowth startup just showed me their A/B results for customer satisfaction across various speech-to-text engines.
Fine-tuned Nvidia Parakeet (on-device) is smoking Gemini and Deepgram out of the water in this case.
Bullish on fine-tuned open-source models. You CAN differentiate in speech-to-text if you decide to care and move away from 3rd-party model APIs.
argmax@argmax
Customize Speech-to-text for Healthcare (in real-time) Transcribing medical conversations requires systems that continually adapt to the newly developed and approved medications, tests, and procedures. Furthermore, there are more than 135 medical specialties, each bringing its unique vocabulary to learn. General-purpose systems are simply not useful in these settings. A popular method for continual adaptation is to fine-tune general-purpose speech-to-text models on evolving vocabularies. However, this requires frequent production deployments with significant updates, potentially leading to excessive time-to-market delays and engineering overhead. The newly improved Argmax Custom Vocabulary feature enables developers to customize speech-to-text in real-time in a self-serve fashion: - Updating the system vocabulary is a configuration change, not a model or system update. - Each medical specialty can easily configure its unique vocabulary to scalably customize behavior in a fine-grained fashion. - Accuracy surpasses medical-domain fine-tuned models in many cases, thanks to precision-targeted vocabularies. (Numbers in replies) As a concrete example, here is how Argmax performs on a file that vocalizes all medications approved by the FDA in 2025.
English
A violinist played for 45 minutes in the Washington D.C. subway.
Of the 1,097 people who passed by, seven stopped to listen to him, and one recognized him.
He collected $32.17 in tips from 27 passersby (excluding $20 from the one who recognized him).
Only one person knew this, but the violinist was Joshua Bell, one of the best musicians in the world. In that subway, Joshua played one of the most intricate pieces ever written with a violin worth 3.5 million dollars.
Two days before he played in the subway, Joshua Bell sold out a Boston theatre, and the seats averaged about $100.
The experiment proved that the extraordinary in an ordinary environment does not shine and is so often overlooked and undervalued.
There are brilliantly talented people everywhere who aren’t receiving the recognition and reward they deserve. But once they arm themselves with value and confidence and remove themselves from an environment that isn’t serving them, they thrive and grow.
Your gut is telling you something. Listen to it if it’s telling you where you are isn’t enough!
Go where you are appreciated and valued.
Know Your Worth.
Credit : @foundconsciousness via IG
English
Reply to this tweet with "AWS" and we’ll tell you which AWS Service you are
English
@ai_for_success Pretty much what Sutton said in that Dwarkesh podcast.
English
Demis Hassabis: Today’s AI is still frozen after training, unable to continuously learn from real world experience, which blocks true general intelligence.
English
There are stuff in real life that cannot be transmitted in books and stuff in books you can never pick up in real life.
The disease is the inclination to learn from books what you must learn in real life and attempts to reinvent from experience what can only be found in books.
English
@atiorh @ClementDelangue Can you explain a bit about the custom vocab and the 0.6 B param model associated with it?
English
Open secret: Frontier generalist models sound cool but specialized models are the workhorse of the industry.
The product below:
- 0.6b for speech-to-text
- 0.1b for speakers
- 0.6b for custom vocabulary
3 specialized models totaling 1.3b parameters running in real-time on 6-year-old entry-level hardware at imperceptible load to the system. Oh and accuracy beats frontier generalists.
argmax@argmax
Introducing Real-time Transcription with Speakers! - Step change in accuracy, surpassing top cloud APIs - Faster than real-time on Mac and iPhone - Still under 3 watts when all features are enabled Available in Argmax SDK 2.0 for early access! Benchmarks and details in comments.
English
@jeremyphoward Once heard a signal processing course professor extoll Tukey and that is it and haven’t heard anyone bring him up.
English
Why isn't Tukey more well-known? He's the godfather of data science.
Coined the terms "bit", "software", and "exploratory data analysis". Made FFTs usable. And much more…
en.wikipedia.org/wiki/John_Tukey
English
@srush_nlp Curious why. In my understanding in the world of speech, folks down sample a high dim 16000 Hz signal to 12 Hz audio tokens using neural codecs and maybe tricks like that creates a situation where you both lose data and yet your mixed modality dataset is not as large-as text only
English
Think about this talk a lot. There was a time when people were bullish on "feed all the modalities to the LLM," but it didn't really pan out as I would have expected. The discrete / continuous divide remains a interesting challenge in deep learning.
Conference on Language Modeling@COLM_conf
COLM Keynotes: Luke Zettlemoyer Mixed-modal Language Modeling youtu.be/PdsKNtEofFY
English
Highly irresponsible, malicious tweet. The report clearly clarifies that medical professionals said it is because of garmi and ghabrahat, but all vitals are okay. These girls will now thankfully never get HPV or cervical cancer… but anyway I thought the concern was autism?
Sridhar Vembu@svembu
This is the news item dated Oct 17 2025 from Bihar. Translation from Hindi (Google Translate): "Saharsa: 30-35 female students fainted after receiving the HPV vaccine...They were admitted to Sadar Hospital...The District Magistrate declared the situation normal." Facts: 1. HPV is sexually transmitted. It does not transmit via regular contact or even hugging. 2. We are giving it to very young girls in India. 3. News from Tamil Nadu just 3 days ago: "Tamil Nadu Paves the Way for a Cancer-Free Future With Free HPV Vaccines" This is NOT a political issue. Please do NOT make this potlicial. Most states in India are doing it and our political leaders across the spectrum have been told that ever more vaccine shots are in the best interest of Indians. I do not blame the politicians because when someone like me can be shouted down so vocally, it would be political suicide for a politician (of any party) to talk about, particularly when the mainstream consensus is so strongly in favor of ever more vaccine shots. It does not help our business or my "image" to talk about this. Even my friends advise me not to talk about it because they are worried for my safety or my image. I have no hidden agenda here. I will state what I believe again: we are giving way too many vaccines to young children. This does NOT mean we should stop ALL vaccines. So let the critics not twist my words. Do not let the doctors who think I am science-illiterate shout this down. These voices must be heard.
English
@doodlestein @simonw Why not? They could have included this in test, no? Unless I am wrong in assuming the popularity of the pelican test. Any popular sniffs test can be scooped by big labs. Is there anything else that makes this hard to game? I am trying to learn.
English
@simonw Is it weird that I trust your pelican test more than all the official fancy benchmarks? It’s the one test you can be reasonably confident they didn’t game/manipulate and train on test.
English
Whoa. HUGE new coding specialist model from Qwen, accompanied by their own fork of gemini-cli (qwen-code) and some very impressive results on the various coding benchmarks
It drew me an OK pelican
Qwen@Alibaba_Qwen
>>> Qwen3-Coder is here! ✅ We’re releasing Qwen3-Coder-480B-A35B-Instruct, our most powerful open agentic code model to date. This 480B-parameter Mixture-of-Experts model (35B active) natively supports 256K context and scales to 1M context with extrapolation. It achieves top-tier performance across multiple agentic coding benchmarks among open models, including SWE-bench-Verified!!! 🚀 Alongside the model, we're also open-sourcing a command-line tool for agentic coding: Qwen Code. Forked from Gemini Code, it includes custom prompts and function call protocols to fully unlock Qwen3-Coder’s capabilities. Qwen3-Coder works seamlessly with the community’s best developer tools. As a foundation model, we hope it can be used anywhere across the digital world — Agentic Coding in the World! 💬 Chat: chat.qwen.ai 📚 Blog: qwenlm.github.io/blog/qwen3-cod… 🤗 Model: hf.co/Qwen/Qwen3-Cod… 🤖 Qwen Code: github.com/QwenLM/qwen-co…
English
@yacineMTB Execution and distribution is everything. Most will not make a penny. You may make some spare change.
English
I'm benefiting greatly from google's open source software yet again, and honestly, I'm not even sure it's a good idea for them to open source this much of the sauce. The value that I can just.. git clone into my server.. it's astounding. It's worth millions of dollars to me
English
@Behn_lstn @hamptonism bare python - your closeness to production vs academia will influence you preference
English
@RuskEating @hamptonism What language do you use to run sims? My stats professor insists we use R but I prefer python.
English
@Behn_lstn @hamptonism Interestingly I too did a mont carlo sim but implicitly introduced an under counting bug that implicitly considered only cases where juliet arrives after romeo. Here you get exactly half the correct answer.
English
@hamptonism A bunch of people were saying 7/16 so I decided to do a Monte Carlo sim and it turns out I was wrong. I found a good visual explanation for the correct answer that makes it more intuitive hf🙃 math.stackexchange.com/questions/1279…
English
A little bit of update from me: I will join the awesome team at @AnthropicAI in two weeks.
English