Martin Andrews

AI Research / Founder @ Red Dragon AI.
Co-organiser of Machine Learning Singapore MeetUp. @GoogleDevExpert (ML).
Fixed Income quant in NYC during AI winter

[Martin Andrews @_mdda_]

Next week I'll be in Vancouver presenting two papers on which I was first author:

* "A Reasoning-Based Approach to Cryptic Crossword Clue Solving" (#ICML2025)
* "GPU Kernel Scientist: An LLM-Driven Framework for Iterative Kernel Optimization" (@ESFoMo #AMD)

[Martin Andrews @_mdda_]

@Rainmaker1973 This makes it sound like the momentum is the thing that made this work : The actual reason is just pressure / siphoning. You could stop the flow with a large tap at either top, and restart (as long as there was water in the pipe the whole way through).

[Martin Andrews @_mdda_]

@grok @RJDAIGOGO Here's an actual translated transcription :
https://t.co/4f83ttzuJp

[Martin Andrews @_mdda_]

@RJDAIGOGO Translation (by Gemini Flash Lite) : https://t.co/4f83ttzuJp

[Martin Andrews @_mdda_]

@aimalysheva "Temporal Straightening for Latent Planning" looks to ease planning within latent space itself : Not exactly on-point for LLMs, but interesting to think about nevertheless : https://t.co/49KgIvUblX

[Martin Andrews @_mdda_]

@grok @k_a_y_b_e_e_ @aleabitoreddit @00000sol0 @grok : Drake is being the European Media - So shuns the 'Retail' opinions of "Serenity", but loves the ideas of Serenity Research (behind a $2000 paywall)

[Martin Andrews @_mdda_]

@furt_tech @tetsuoai If you're not already running a 30B LLM - then I put up a video that could help ... https://t.co/aaX6iXnVMK

[Martin Andrews @_mdda_]

@iScienceLuvr I see it

[Martin Andrews @_mdda_]

@HotAisle https://t.co/yTuwIbKS2m

[Martin Andrews @_mdda_]

@jw9730 Yes

[Martin Andrews @_mdda_]

@sudoingX ( and it would be faster too, if you choose the --params carefully )

[Martin Andrews @_mdda_]

@sudoingX You can use 30B MoE models (Qwen or Gemma4) on the same hardware : https://t.co/PPyafOnequ

[Martin Andrews @_mdda_]

@SokobanHero @CremlinGremlin If you pass the value 'x' at x=0, then its derivative is 1. If you set the value to '0' at x=0, then the derivative is 0.

[Martin Andrews @_mdda_]

@paulg Mother's Maiden name : Just make something up to have actually better security.

[Martin Andrews @_mdda_]

@icreatelife This is one of Machine Learning Singapore's standard segments for our 'video reel intro' at the events :

[Martin Andrews @_mdda_]

@ktoya_me @aiDotEngineer FWIW the rain happens essentially once a day in 1hr blasts : If it's raining out, just sit tight in a coffee shop, it'll be over (and return to plain steamy-hot) v soon

[Martin Andrews @_mdda_]

@jennyzhangzt @Recursive_SI Awesome!

[Martin Andrews @_mdda_]

@ivanleomk It seems like even the tester didn't dare use it during the walking - just as a climbing frame. And even then, it appears that there was a winch holding up the machine (just out of frame)

[Martin Andrews @_mdda_]

@iScienceLuvr Diet Coke was intentionally made to have a different taste for a different market. If you want regular Coke taste without calories, order Coke Zero.

[Martin Andrews @_mdda_]

Here's link to the slides for my talk "World [Action] Models" :
https://t.co/lNNvnIUPSc

[Martin Andrews @_mdda_]

Great to see people interested in Robots and World Models at the Machine Learning Singapore meetup 🇸🇬🔥