Postdoc @IHUStrasbourg | AI for Surgical Safety & Interpretable ML | PhD in Cardiac Surgery @UniHeidelberg | Robust CV for surgical workflows | 🇮🇳→🇩🇪→🇫🇷
Strasbourg, France
Joined June 2019
- Tweets265
- Following1.7K
- Followers438
- Likes14.7K
lalith retweeted
The hardest problems are rarely solved by adding more complexity to the solution -- they are solved by reframing the question until a simpler, clearer answer reveals itself.
antiquity: you are written in the stars
the enlightenment: you are god’s clockwork
the computer age: you are elegant software
and now:
The aviation-surgery comparison is useful but probably not explored fully. Aviation's safety systems sit on a mature sensor stack. In MIS, the endoscopic video stream does most of the heavy lifting, and building safety systems native to that modality is a very different problem.
Some thoughts here: linkedin.com/posts/lalithnag…
lalith retweeted
"You are a professional translation assistant. Detect the source language automatically. Translate the user's text into English. Preserve tone, meaning, punctuation, emoji, and inline formatting. Return only the translated text without commentary, labels, or quotes."
They just yolo-ed the prompt into GPT-5 model and called it a day 😂
pov you’re a squirrel watching sutton on dwarkesh: field notes on the persistent anthropocentric bias in reinforcement learning
.@RichardSSutton, father of reinforcement learning, doesn’t think LLMs are bitter-lesson-pilled.
My steel man of Richard’s position: we need some new architecture to enable continual (on-the-job) learning.
And if we have continual learning, we don't need a special training phase - the agent just learns on-the-fly - like all humans, and indeed, like all animals.
This new paradigm will render our current approach with LLMs obsolete.
I did my best to represent the view that LLMs will function as the foundation on which this experiential learning can happen. Some sparks flew.
0:00:00 – Are LLMs a dead-end?
0:13:51 – Do humans do imitation learning?
0:23:57 – The Era of Experience
0:34:25 – Current architectures generalize poorly out of distribution
0:42:17 – Surprises in the AI field
0:47:28 – Will The Bitter Lesson still apply after AGI?
0:54:35 – Succession to AI
we are so back
OpenAI, SAP & Microsoft are launching OpenAI for Germany—a partnership to bring frontier AI to Germany’s public sector, through a sovereign, certified cloud environment.
Built on SAP’s Delos Cloud and running on Microsoft Azure, this new initiative will help employees across German governments, administrations and research institutions use our technology to spend more time on people, not paperwork.
openai.com/global-affairs/op…
funny, I don’t see the actual CV researcher categories here:
‘fix my loss function so it doesn’t collapse in 2 epochs’ (40%)
‘explain why my GPU dies at batch size = 1’ (35%)
‘convince me my architecture isn’t just ResNet in makeup’ (25%)
This Post is from a suspended account. Learn more
gaussian splats making it to Superman before half of computer vision Twitter figured out what they were
It's a bird. It's a plane. It's gaussian splatting.
When @Framestore needed to bring Kryptonian tech to Earth, they turned to gaussian splatting. In the recently released Superman, every shot of Kal-El's parents is a dynamic splat.
I spoke to the team at Framestore about capturing the splats at @8Infinite8 and how they used CG Nomads GSOPs in @sidefx's Houdini to help make the highest profile public use of gaussian splatting to date.
Article
radiancefields.com/gaussian-…
C’mon man, you can’t start a percentage graph at 50 @AnthropicAI
Today we're releasing Claude Opus 4.1, an upgrade to Claude Opus 4 on agentic tasks, real-world coding, and reasoning.
ALT Bar chart comparing software engineering accuracy between two versions of Opus 4. The first bar shows Opus 4 (May 2025) at 72.5% accuracy and the second bar shows Opus 4.1 (Aug 2025) at 74.5% accuracy.
Now would be a good time to read/watch ‘A scanner darkly’
Japanese designer Kunihiko Morinaga unveiled a collection made from LED-embedded textiles that display up to 60 different patterns—on demand. Soft, flexible, and fully programmable, these pieces mark a turning point where fashion meets interactive tech.
#technews