Martin @sub31

“Simplicity is the ultimate sophistication.” Leonardo da Vinci

[Martin @sub31 @sub31]

@fchollet yes - but much harder to do? first principle type approach likely the way to go. reminds me of the #gibberlink phenomenon. leave it to the GPT/RL and we'll automate processes which work, but where the steps are made up to provide rationale post hoc.

[Martin @sub31 @sub31]

starting w top level categories for #mappingai2025. i found something to use as a strawman via @scgupta of https://t.co/hsOywK2nz1. i would like to get clear about what's being worked on & how it fits together. is this about right? suggestions/comments? pls retweet.

[Martin @sub31 @sub31]

i am thinking about a concept map for #deeplearning and #aireasoning - wondering if this has been done (and to what extent) already... anyone know? #mappingai

[oliver caviglioli @olicav]

I read two articles in the current edition of Wired on artificial intelligence (AI) and wanted to integrate them, so drew this concept map of the ideas contained in them.

[Martin @sub31 @sub31]

so i'm wondering if the approach that @deepseek_ai came via reverse engineering @openai's o1 model - or were they on that path already? hat tip to them for writing it up so well and making it #opensourcesoftware

[Roberta Raileanu @robertarail]

Turns out we only needed a better base model and to scale up RL on a diverse set of tasks.

We even applied RL directly on top of the base model without SFT, like DeepSeek-R1.

Thanks @deepseek_ai for the detailed technical report.

[Martin @sub31 @sub31]

blog post on Lecun's IJEPA from the interesting looking #aishop @encord_team: https://t.co/JUB8Zb977Y

[Martin @sub31 @sub31]

#agibreakthrough imminent? do we have an LLM which can 'generalise outside its training distribution' (per @daveshapi)? i have been doing some exploring via @perplexity_ai into the zoomed out context - pdf to share here: https://t.co/7ORnV34150 - not for the pros!

[Martin @sub31 @sub31]

#worldmodels - Lecun vs Schmidhuber: https://t.co/nndFJNgszD

[Noam Brown @polynoamial]

It can be hard to “feel the AGI” until you see an AI surpass top humans in a domain you care deeply about. Competitive coders will feel it within a couple years. Paul is early but I think writers will feel it too. Everyone will have their Lee Sedol moment at a different time.

[Machine Learning Street Talk @MLStreetTalk]

"o3 is the alexnet moment for program synthesis" - today is a pretty big day folks. We need to update our priors. o3 is zero-shotting extremely high scores on ARC-AGI which is a step towards the spirit of "developer aware generalisation" which Chollet originally envisioned.

[Martin @sub31 @sub31]

@Se_Railway I'm a regular user and love some of the deals you offer, but the #userexperience of your website/apps - and comms could do with attention. Happy to help! Spot the difficulty below...

[Martin @sub31 @sub31]

Very buggy #tagging functionality from @Bitly - #uxfail, company in trouble?

[Martin @sub31 @sub31]

#primaryright or #primaryleft - food for thought on button placement in #uidesign from @ArtemSyzonenko (who is from Dnipro in Ukraine) 🇺🇦 - best wishes to Artem... https://t.co/jz9lJ3afCH

[Martin @sub31 @sub31]

Even GDS is telling GDS to shut down Verify https://t.co/d0Y7Nq6OMH via @computerweekly

[Martin @sub31 @sub31]

Interesting - wonder how the analytics pan out for this: #readthearticlebeforesharing #nudgetheory

[Martin @sub31 @sub31]

@figurativeapp I see the list of files in my project folder, but when I tap - nothing happens. it is highlighted in blue but does not open. Using ipad mini 5th gen, ios 14.4.2 - any ideas? Thanks.

[Simon Wardley @swardley]

A staggering and unimaginable cost with no clear evidence of effectiveness ... afaik, that's common language in the private sector for projects run by ex McKinsey consultants.

Test and trace should have been built by the NHS working with local councils - https://t.co/TDR88vcDsP

[Martin @sub31 @sub31]

WhatsApp alternatives and basic #identitysecurity from @guardian
https://t.co/R6jZcWaR6g