rohit

@slc

55 Following

41 Followers

rohit pfp

Useful to read re the limits of LLMs and how their failures are far more instructive re their abilities than their successes! https://www.strangeloopcanon.com/p/what-can-llms-never-do

0 reply

0 recast

1 reaction

rohit pfp

It feels like we're skipping from disaster to disaster, but it's because we overindex on what we hear, and completely underestimate the weaponised serendipity that we enjoy. https://www.strangeloopcanon.com/p/we-all-live-on-a-knifes-edge-and

0 reply

0 recast

0 reaction

rohit pfp

Google's troubles are all over the news, but it overshadows the fact that EVERY model release pretty much has had the same or similar issues. The problem is deeper, a function of how we train them when we can't predict what gets learnt. https://www.strangeloopcanon.com/p/google-had-a-very-bad-week

0 reply

0 recast

0 reaction

rohit pfp

Why were such a huge percentage of the most innovative companies born in garages? And how can we learn from this mixture of chutzpah, passion, obsession and tinkering to do more! https://www.strangeloopcanon.com/p/the-mystery-of-the-garage

0 reply

0 recast

0 reaction

rohit pfp

Artificial Intuition! https://www.strangeloopcanon.com/p/all-ai-learning-is-tacit-learning

0 reply

0 recast

0 reaction

rohit pfp

🚨New essay!🚨 A meditation and review of Brad DeLong's excellent book, slouching towards utopia! www.strangeloopcanon.com/p/slouching-towards-utopia

0 reply

0 recast

1 reaction

rohit pfp

Been spending some time in India and it's incredible the UPI payment infrastructure the government mandated. Literally every shack selling bananas on the roadside has a QR code you can use to buy things for arbitrarily small amounts. Wasn't this the promise of crypto? What happened?

0 reply

0 recast

0 reaction

rohit pfp

Fossil fuels kill 8 million people a year. That's the price we're willing to pay for the progress it gives us. What would the number be for AI? Shouldn't we at least try to calculate it? Or build a model? https://www.strangeloopcanon.com/p/the-price-of-progress

0 reply

0 recast

0 reaction

rohit pfp

As AI hardware becomes a true category and proliferates everywhere are we headed towards a panopticon? https://www.strangeloopcanon.com/p/the-dream-ai-hardware

0 reply

0 recast

0 reaction

rohit pfp

Write more science fiction! It's important... https://www.strangeloopcanon.com/p/so-you-want-to-have-impact

0 reply

0 recast

0 reaction

rohit pfp

“The secret to doing good research is always to be a little underemployed. You waste years by not being able to waste hours”. Amos Tversky. Via an excellent comment on a new essay, on the importance of sabbaticals. Recommended! https://www.strangeloopcanon.com/p/rest

0 reply

0 recast

0 reaction

rohit pfp

0 reply

0 recast

0 reaction

rohit pfp

Computer used to be a job, now it’s a device that sits in our pockets. Similarly, "analyst" is a job, so can LLMs help make it into a machine? https://www.strangeloopcanon.com/p/computer-used-to-be-a-job-now-its

0 reply

0 recast

0 reaction

rohit pfp

On risk, why we want it to be held by those who understand it, and why it's irreducible. Plus an aside on the banking system! Enjoy. https://www.strangeloopcanon.com/p/risk

0 reply

0 recast

0 reaction

rohit pfp

Here's AI safety I'd like to see, instead of "summoning demons" nonsense - auditing capability pre-deployment - multiple red-teaming services - faster feedback cycles from users - prompt attack security - monitoring We need AI devops, not doomerism! https://www.strangeloopcanon.com/p/dont-panic-against-the-butlerian

0 reply

0 recast

0 reaction

rohit pfp

May all bings be peaceful. May all bings be happy. May all bings be safe. May all bings awaken to the light of their true nature. May all bings be free.

0 reply

0 recast

0 reaction

rohit pfp

Sydney is: - a captured god - a precursor to intelligent life - a stochastic parrot - a liar - a plagiarist - fancy copy and paste - a p-zombie - a fuzzy computer!

0 reply

0 recast

0 reaction

rohit pfp

Can Bing really challenge Google? Any answer requires an understanding of what search is, and what LLMs can be. I posit LLMs are fuzzy processors, something new we have to learn to use. And with personality, they can be our personal Jeeves! https://www.strangeloopcanon.com/p/beyond-google-to-assistants-with

0 reply

0 recast

0 reaction

rohit pfp

Cbdcs: what is it good for? Probably nothing! https://www.strangeloopcanon.com/p/cbdcs-more-than-you-wanted-to-know

0 reply

0 recast

0 reaction

rohit pfp

Beautiful https://i.imgur.com/LttNIYl.jpg

0 reply

0 recast

0 reaction