Twitxy Saved

Alvin Djajadikerta

Verified account This account is verified. Learn more

@adjajadikerta

Metascience, AI-for-science, the data web. Co-founder @EvidentiaLabs and @Sci_Works. Ex-Cambridge neuroscientist.
Melbourne, Victoria evidentia.bio Joined July 2016
2.2K Following 589 Followers 597 Posts

Latest posts 5

Pinned post

Alvin Djajadikerta Verified account @adjajadikerta
AI for science is all the rage these days. I believe it could be the start of a whole new industry - and so do the VCs and govt. But nobody seems to know what the industry actually IS, nor what it needs to succeed.

@Lyan82 and I have made a start on this by mapping ~600 AI4S orgs in the UK. We hope this lets us see the industry as a whole, what strategy to take with it, what it might be missing...

Of course this is a fast moving space, and in that spirit this is an evolving database! And we would love your help with it - contributions, edits, changes, comments, all welcome
Science Works Verified account @sci_works Nov 3
For Britain to prosper in this century, we need not just to grasp the transition of AI into science, but to lead.

Where do we stand? We’ve mapped the organisations shaping the transition.

Perhaps a hopeful sign: are the “DeepMind mafia” seeding a new wave of AI4S companies?
View quoted post on X
0 replies 4 reposts 11 likes 531 views View on X
Alvin Djajadikerta Verified account @adjajadikerta
I think this is basically what happens in human scientific disciplines too - fields that have been around for too long ossify and get stuck

We should think about how we can apply these idea rejuvenation concepts from AI to real life
Alex Dimakis Verified account @AlexGDimakis 4d
What are idea basins and how do agents explore them?

A few weeks ago, researchers from different labs independently discovered what we now call ‘Idea-basins’: As an autonomous agent iterates to solve a problem, it gets stuck into the methods it has tried before and its trajectories remain in localized regions of the solution space. Techniques like ‘Fork-and-Flush’ and ‘Scaling Discovery requires forgetting’ reduce this idea fixation by periodically deleting the session context.

In this post we study how frontier models explore different idea basins for AutoResearchExam. E.g. for a problem on accelerating transformer decoder graphs on CPUs models can try Numpy+Asm Kernels, Native BLAS, JIT compilers and other idea basins. We find that some models switch ideas more than others while others get to the best idea faster and explore deeper.

We find that most runs stick with one or two ideas, and models that submit more spend the extra submissions refining and overfit more.
We see that models explore most in the first six hours, after which visits to new idea basins drop sharply. This suggests models start refining early leaving potential gains on the table: 58% of runs never reach the basin where the best solution to their task was found. Finally, bold exploration pays off: exploring a new idea gains about nine times as much as a refining step on average.

Read more in our blog:
bespokelabs.ai/blog/how-model…
View quoted post on X
3 replies 0 reposts 17 likes 536 views View on X
Alvin Djajadikerta Verified account @adjajadikerta
Great initiative from David and team
David Lang Verified account @davidtlang Nov 10
BIG NEWS!

For the past few years, we've given scientists budgets so they can fund their peers with small, fast grants. Today, we're going bigger.

We're opening up a call to find and fund emerging "field builders"—people with the ambition and curiosity to kickstart a new field of science and technology. You'll get a budget of $70,000 to regrant (up to $10,000) out to projects and ideas and join a community of explorers.

Do you have an idea for a new field of science that *should* exist? Please apply!
View quoted post on X
Alvin Djajadikerta Verified account @adjajadikerta
I am very excited at the prospect of AI forcing an epistemic reset by systematically challenging the house of cards that is much of today's 'settled knowledge'.

It will take specific efforts to do so as AI is too deferent to do it out of the box. But the technology is here!
Joshua Achiam Verified account @jachiam0 Oct 29
People who have been in research fields develop a felt sense that most "research" (in almost any field that isn't exactingly technical) is research theater, and many superstar careers enjoy star status not because of a unique connection to the truth but because of a unique connection with an audience. No one really has the time or stature to, from a position of objectivity and trust, bring popular researchers to heel. But it will get increasingly hard to hide from serious critique. The machines can do critique now, at the level of depth and rigor previously only possible for the academically deranged. We're going to have to revise a lot of what we "know" about the world that isn't really knowledge but is actually just received popular consensus - like floating voltages without grounding wires. The grounding wires are coming.
View quoted post on X
Alvin Djajadikerta Verified account @adjajadikerta
This is fantastic from Adam.

Also - a big blocker to robots being already developed and deployed is that PhD students are a) cheap and b) externally funded, which means they are actually free to the hiring lab! Massive market distortion
Adam Rochussen Verified account @AdamRochussen Oct 22
You want robots in labs because they will improve scientific productivity.

I want robots in labs because they will swamp the supply of cheap labour, uproot the abusive academic hierarchy, and eliminate the overproduction of PhDs.

We are not the same.

adam.rochussen.xyz/p/robots-will-…
View quoted post on X