Skip to content

ViralLoop

Your next post already worked.

←

Write from this board Add posts Share

AllXLinkedInInstagramTikTok

FilterOrderGridTableSelect

Loading the board

Stride — quantify or don't post

Remix this board

24 posts · curated by Nam Nguyen

Performance positioning for an audience that already measures everything. A measurable claim with the mechanism behind it, and no hedging anywhere in the sentence.

AllX LinkedIn Instagram TikTok

Sebastien Bubeck@SebastienBubeck

Claim: gpt-5-pro can prove new interesting mathematics. Proof: I took a convex optimization paper with a clean open problem in it and asked gpt-5-pro to work on it. It proved a better bound than what is in the paper, and I checked the proof it's correct. Details below.

Opens with an observation

8K likes308 repliesAIX1y ago

Dr. Rhonda Patrick@foundmyfitness

Exercise may mitigate some of the mortality risks linked to poor sleep. People who slept less than 6 hours a night had a decreased life expectancy compared to those that slept between 6-8 hours a night - but only in people that were NOT physically active.

Opens with an observation

2.6K likes42 repliesHealthX3y ago

Dr. Rhonda Patrick@foundmyfitness

Getting high-quality sleep protects the brain. A new study shows that getting fewer than 5 hours of sleep per night may double the risk of developing dementia. Sleep activates the glymphatic system which cleans the brain of toxic metabolites and debris. t.co

Opens with an observation

1.2K likes27 repliesHealthX5y ago

Satchin Panda@SatchinPanda

NYTimes oversimplified this story and it is a terrible message for the public. All TRE studies reduce eating from 12+ to 8 h. In this study, People with 10 h eating were asked to eat within 8h. There is no precedence that reducing eating from 10 to 8 h will show any benefits.

Opens with an observation

292 likes18 repliesScienceX4y ago

Simpler way to understand DeepSeek weren't lying about 50k H100s or training costs for V3/R1 We have the model, its a 35b active, 640b Mixture of Experts We know that spec is 2-3m hours to train Models get worse with more compute after a certain point! t.co

Opens with a number

255 likes11 repliesAIX1y ago

Jaime Sevilla@Jsevillamol

DeepSeek-V3 is impressively efficient. It has 37B active params and was pre-trained on 14.8T tokens, amounting to 6 x 37B x 14.8T = 3e24 FLOP. That's 10x less compute than llama 3.1 430B, yet better benchmark performance.

Opens with an observation

151 likes4 repliesAIX1y ago

Qwen 3.5 Small models - fully open source - beats models 4x it’s size - 9B model performs on par with GPT OSS 120B while being 13x smaller - outperforms Gemini 3 flash and Claude sonnet 4.5 on select benchmarks - runs on any laptop - even works on a phone - completely free.

Opens with an observation

6.8K likes124 repliesAIX6mo ago

Aaron Ng@localghost

Here’s Deepseek r1 1.5B thinking through a problem — it’s comparable to 4o and Claude 3.5 Sonnet in a number of domains like math. Except… it’s a 1.5B model… and can run on virtually any hardware. Truly a huge efficiency leap.

Opens with a number

3.5K likes94 repliesAIX1y ago

The ex-quants in China dropped the best open-source LLM in the world: DeepSeek V3! — 671B params, 37B active MoE — On par with 3.5 Sonnet and 4o — $0.27/Mtok input, $1.1/Mtok output — 60tok/s — 128k context — Trained with just $5.5M on 14.8T tok! 53-page technical paper is GOLD

Opens with an observation

3.2K likes68 repliesAIX1y ago

VideoLAN@videolan

About the "security issue" on #VLC : VLC is not vulnerable. tl;dr: the issue is in a 3rd party library, called libebml, which was fixed more than 16 months ago. VLC since version 3.0.3 has the correct version shipped, and @MITREcorp did not even check their claim. Thread:

Opens with an observation

3.1K likes45 repliesProductX7y ago

JustDario@DarioCpx

The news just broke that SoftBank could not secure a 6bn$ vs its OpenAI shares Let me tell you a thing: if you cannot secure a 6bn$ loan against collateral you claim is worth ~100bn$, then the latter isn't worth ~100bn$. In this case, it might be worth not much more than 6bn$.

Opens with an observation

2.9K likes68 repliesAIX3mo ago

ClaudeDevs@ClaudeDevs

Before, running a remote MCP server meant managing session state, which limited where you could run it. Now that MCP is stateless, you can deploy on serverless and edge infrastructure, or scale horizontally behind any load balancer.

Opens with an observation

2.4K likes48 repliesEngineeringX1mo ago

Nabeel S. Qureshi@nabeelqu

Everyone is way overindexing on the $5.5m final training run number from DeepSeek. - GPU capex probably $1BN+ - Running costs are probably $X00M+/year - ~150 top-tier authors on the v3 technical paper, $50m+/year They're not some ragtag outfit, this was a huge operation.

Opens with a number

1.8K likes62 repliesHealthX1y ago

Jan Leike@janleike

New research result: we use Claude to make fully autonomous progress on scalable oversight research, as measured by performance gap recovered (PGR). Claude iterates on a number of different techniques and ends up significantly outperforming human researchers for $18k in credits.

Opens with an observation

1.3K likes40 repliesAIX5mo ago

Yuchen Jin@Yuchenj_UW

Surprised to see the Gemini app hit 650M monthly active users, thanks to Nano Banana I guess. ChatGPT has 800M weekly users (so likely ~1B monthly). Google’s catching up fast on the consumer app side. Gemini’s already generating 7B tokens/min vs. OpenAI’s 6B tokens/min?

Opens with a number

743 likes57 repliesAIX10mo ago

Matt Shumer@mattshumer_

There's a new king of open-source. The new LLM from @databricks just beat Mixtral! - 132B total params (16 experts), 36B active (4 experts) - Trained on *12T* tokens - 32K context length From my initial tests, I'm really impressed.

Opens with an observation

456 likes8 repliesAIX2y ago

Sam Blau@SamMBlau

The Open Molecules 2025 dataset is out! With >100M gold-standard ωB97M-V/def2-TZVPD calcs of biomolecules, electrolytes, metal complexes, and small molecules, OMol is by far the largest, most diverse, and highest quality molecular DFT dataset for training MLIPs ever made 1/N

Opens with a number

422 likes4 repliesHealthX1y ago

OpenRouter is now serving 1.5 quadrillion tokens/yr! That token run rate is: — 15-30% of Google APIs — 20-40% of OpenAI — >50% of Microsoft Azure Foundry That's 15x larger than when we invested a year ago. Revenue has already doubled since this $1.3B round was done in Feb!

Opens with an observation

290 likes33 repliesAIX3mo ago

Oleksii Kuchaiev@kuchaev

Nemotron 3 Super is here — 120B total / 12B active, Hybrid SSM Latent MoE, designed for Blackwell. Truly open: permissive license, open data, open training infra. See analysis on @ArtificialAnlys Details in thread 🧵below:

Opens with a number

275 likes10 repliesHealthX6mo ago

Aran Komatsuzaki@arankomatsuzaki

Apple presents OpenELM - An efficient LM family with open-source training and inference framework - Performs on par with OLMo while requiring 2x fewer pre-training tokens repo: t.co hf: t.co abs: t.co

Opens with an observation

233 likes5 repliesAIX2y ago

Chunky beast, needs 320 Gb VRAM likely 4 bit, likely is being run 8 bit on 8 x 80 Gb GPUs. As 25% of weights active any given time runs at 70b LLaMA 2 speeds. MoE models tail off larger so this is very interesting to test to see tradeoffs.

Opens with a number

229 likes9 repliesAIX2y ago

Sully@SullyOmarr

the only let down of claude 3.7 is we didn't get any price reductions it's my go to model, but its slowly getting harder to justify the price (especially once you add reasoning tokens) for comparison gemini 2.0 is 30x cheaper @ 0.1m input and 0.4 output

Opens with an observation

198 likes36 repliesAIX1y ago

Alexia Jolicoeur-Martineau@jm_alexia

Claim from the abstract: "106B-parameter MoE (12B active) trained with large-scale reinforcement learning on our end-to-end RL infrastructure stack." I expected all RL from scratch. Reality: Already existing base model + SFT + RL 😿

Opens with an observation

189 likes13 repliesAIX9mo ago

PJ Ace@PJaccetturo

What if Studio Ghibli directed Lord of the Rings? I spent $250 in Kling credits and 9 hours re-editing the Fellowship trailer to bring that vision to life—and I’ll show you exactly how I did it 👇🏼

Opens with a question

93K likes3.8K repliesCreativeX1y ago

Boards are how ViralLoop learns your taste — save posts you admire, and turn the collection into content that sounds like you.

Start your own board

Stride — quantify or don't post · ViralLoop