Andrew Sellergren

Triathlete | Coach | Software Engineer @Google

[Andrew Sellergren @asellerg]

Download it on HuggingFace now!

https://t.co/wuHZNO1cEH

[Sundar Pichai @sundarpichai]

MedGemma 1.5 is a major upgrade to our open models for healthcare developers.

The new 4B model enables developers to build applications that natively interpret full 3D scans (CTs, MRIs) with high efficiency - a first, we believe, for an open medical generalist model. MedGemma 1.5 also pairs well with MedASR, our speech-to-text model fine-tuned for highly accurate medical dictation.

Developers can now use these multimodal capabilities to build medical apps that reach patients in more places.

[WSOP - World Series of Poker @WSOP]

Negreanu Calls Out His Table!

After some questionable play near the money bubble on Day 2 of the $10,000 2-7 Lowball Championship, Daniel Negreanu managed to survive with just a few big blinds.

@RealKidPoker then doubled up through one of the greatest no-limit hold'em players, @WAFoxen, and Negreanu didn't hold back!

[Andrew Sellergren @asellerg]

@BlackHC He was a big proponent of a preemptive nuclear strike against Russia, so unfortunately this isn't very surprising...

[Andrew Sellergren @asellerg]

@mattshumer_ "Anyone still making that argument either hasn't used the current models, has an incentive to downplay what's happening..."

Unlike you and Dario who definitely don't have the incentive play up the hype?

The last time you went viral it was also overhyping https://t.co/6spPOgtnoM

[Ted Werbel @tedx_ai]

For anyone wondering what happened, here's a quick recap of the Reflection 70b drama:

> Matt Schumer, CEO of HyperWriteAI & OthersideAI, published a model onto HuggingFace and claimed that it had reached higher scores in various benchmarks which would have made his smaller, open source model more capable that some of the biggest models on the market

> Tons of AI researchers, influencers, academics and executives start promoting this all over the web. Integration with OpenRouter goes live. Matt gains ~14k new followers on X and promotes his brands + a synthetic data generation tool that he's invested in.

> Attempts to replicate benchmarks failed miserably - performing worse than many existing models. Matt suggests that this is a glitch, then says he's retraining the model and then goes silent as real AI engineers on X begin dismantling his claims by analyzing the tokenizer (reference the image and links below)

So what did Matt do?!

> He probably wrapped an Anthropic model into an API endpoint and filtered out strings such as 'claude' which would reveal the original source model.

> This was all likely a marketing stunt to increase his follower count and promote businesses he owns or has invested in...

> If this is true, he actually thought he'd get away with it and that no one would test his claims! Perhaps greed and delusions of grandeur have gotten the best of him..

Only time will tell if he can redeem himself and his reputation. Regardless of the outcome, one thing is clear - WE NEED BETTER BENCHMARKS FOR MODELS AND AGENTS!

https://t.co/gGU7ijfBHZ

[Andrew Sellergren @asellerg]

@PokerGOTour @BrynKenney @angryjmonnett @YuriNerdguy Is this being streamed?

[Google AI @GoogleAI]

This week, we released a pair of Gemma updates that developers can use to innovate across the healthcare and translation spaces:

— MedGemma 1.5 4B: Our updated open medical AI model improves performance in 3D imaging and reasoning over medical records. We also kicked off the MedGemma Impact Challenge on @Kaggle to inspire development of AI applications that can help patients
— TranslateGemma: This collection of multimodal open translation models is built on Gemma 3 and enables developers to build translation tools across 55 language pairs. The models are efficiently compact (available in 4B, 12B, and 27B parameter sizes), achieving high-fidelity translation quality with lower latency

You can access both Gemma models on @huggingface, MedGemma on Vertex AI, and TranslateGemma on Kaggle. To learn more about Gemma and our latest breakthroughs from @GoogleDeepMind, visit https://t.co/Mf9nPsX2vA!

[Kaggle @kaggle]

📣 Hackathon Launch Alert! MedGemma Impact Challenge hosted by @GoogleResearch

🎯Build human-centered AI applications by using MedGemma and other open models
💰 $100,000 Prize Pool
⏰ Final Submission: Feb 24, 2026

https://t.co/fwCcQ1B0lx

[Dr. Javier Flores @farmacotips]

“Esta IA acabará con los imagenologos”

Interpreta TAC, RM e histopatología, cruza hallazgos con el contexto clínico y reportes de laboratorio, y razona en términos de enfermedad, no de pixeles.

Si, es el fin, pero del imagenologo saturado.

MedGemma es el futuro de la medicina.

[𝔸𝕦𝕝𝕒𝕤 𝕀𝕟𝕥𝕖𝕝𝕚𝕘𝕖𝕟𝕥𝕖𝕤 @AulasInteligent]

Google lanza MedGemma 1.5 (4B), un LLM open para interpretar imágenes médicas CT, MRI e histopatología, además de RX, dermatologí, fondo de ojo…) con burnos resultados. Y lo acompaña además con MedASR, un modelo abierto de speech-to-text clínico (dictado médico).+

[Omar Sanseviero @osanseviero]

Introducing the MedGemma Impact Challenge

Use MedGemma and other open models to build human-centered applications in the medical space. Over $100k in prizes. Time to build.

https://t.co/W5m0gjon7s

[Jeff Dean @JeffDean]

An updated version of our MedGemma model is out with major accuracy improvements in medical-related tasks. We also released MedASR, which specialized for low error rates for medical speech recognition. We're excited to see what the health community does with these models! 🩺

[Rohan Paul @rohanpaul_ai]

@sundarpichai Google is on a serious roll right now.

[Sundar Pichai @sundarpichai]

MedGemma 1.5 is a major upgrade to our open models for healthcare developers.

The new 4B model enables developers to build applications that natively interpret full 3D scans (CTs, MRIs) with high efficiency - a first, we believe, for an open medical generalist model. MedGemma 1.5 also pairs well with MedASR, our speech-to-text model fine-tuned for highly accurate medical dictation.

Developers can now use these multimodal capabilities to build medical apps that reach patients in more places.

[Google AI Developers @googleaidevs]

Announcing MedGemma 1.5 🩺

This updated open model from @GoogleResearch is small enough to run offline, improves performance on medical imaging use cases, and enhances core capabilities for text, medical records and 2D images.

https://t.co/kVdvdLZapm

[Andrew Sellergren @asellerg]

@davidmbudden Ending this charade with performative highroading is honestly perfect

[Andrew Sellergren @asellerg]

@davidmbudden @PedrusMaximus33 https://t.co/WcZ2gtKLwd

[Budden @davidmbudden]

Also I refuse to acknowledge any lazy copy/paste LLM critiques, unless they come with an explanation why you believe they're incapable of hallucinating false-negatives.

(I'm sure there are bugs in the drafts. I'll get around to fixing them when I'm done partying 🎉)

[Andrew Sellergren @asellerg]

1.6 dE/dt ≤ (2 − A)ΩE. "When A > 1, palinstrophy growth dominates stretching for large Ω, forcing energy depletion near any potential singularity."

How is energy depleted if dE/dt is > 0 (when A = 1.234). At best, you've established a bound on energy growth, but not demonstrated that it is depleted.

[Andrew Sellergren @asellerg]

@TaliaRinger @_p_hacked @goldstein_aa @ezhengs @davidmbudden I did ask around internally when the whole thing started for someone to check on him but didn't get much of a response.

[Andrew Sellergren @asellerg]

@nikitabier @miragemunny Better thing you should solve:

The infinite nonconsensual porn generator running amok in your app

[Andrew Sellergren @asellerg]

@colin_fraser Just a few more hours bro. Just a few more Lean lint errors bro.