https://turntrout.com/why-i-left-google-deepmind
Maybe this is what happens when someone with Jeff Dean's standing tries to quit?
TBH, I'd rather have Jeff Dean working on the creepiest-possible tech for ICE than joining the race to automate AI research. Automating AI research is terrifying.
what why?
I'm curious if that is before or after token costs?
Sometimes I couldn't resist wondering if I'll ever do work that has a tenth of the impact of theirs.
Not a bad combined CV.
Google's advanced AI cannot even exit a mobile app.
I should give it another try…
I don't do huge automatic project wide hands-off agent loops though. I spent a lot of time architecting my systems to be easy to generate code on top of with pointed & detailed prompts. So I'm not abusing context... YMMV
Jeff was a ACM Fellow in 2009 and published the massively influential MapReduce paper in 2004.
I had not read this before, but told many students the same about my PIN code and I a quiz about the last digits. Love it.
Here are some Jeff Dean well sourced facts:
- Already part of engineering of Google indexing systems that lacked basic checksums and ran on non-ECC hardware, allowing silent data corruption.
- One of the authors of LevelDB a database with so many documented crash-consistency, recovery, and data-loss weaknesses for years. Just check their Github project. LevelDB current tracker contains unresolved crash consistency, recovery and corruption reports going back almost 12 years on GitHub
- In AI engineering technical lead, let TensorFlow lose researcher mind share to PyTorch, and caused Google fragmented landscape across TensorFlow and JAX.
- Had the people at Google who invented the Transformer architecture, but failed, to turn that lead into the first dominant public LLM.
- As AI engineering and VP management let Google Brain and DeepMind remain duplicated and internally competitive for too long.
- Let Noam Shazeer leave and then spent heavily to bring him back with nothing to show for.
- Part of Technical VP leadership who had Bard rushed to launch with factual errors in Google own promotional material.
- The first Gemini demonstration overstated how real-time and interactive the system actually was, being basically a fake.
- Part of the VP and AI technical leadership who had Google AI Overviews launched with weak source quality controls and repeated satire and low-quality web content as factual advice.
- Part of teams that launched AlphaChip performance claims that were difficult for outside researchers to reproduce and remain technically disputed.
- Jeff Dean public explanation of Gebru departure was contested and damaged confidence in Google scientific governance.
- Jeff Dean was part of the team at Google that removed or marginalized prominent internal AI ethics critics shortly before many of their warnings became product problems.
- Jeff Dean was one of the managers behind Project Dragonfly supporting censorship.
- Jeff Dean is part of the VP technical leadership approving Project Nimbus supporting an ongoing genocide.
And so does academia. It's just that instead of AI and robotics, PhD students are thrown onto problems that are in large parts slightly tweaked reconfigurations of similar experiments.
Especially in chemistry, biochemistry, material sciences there is a large space of discoveries that are barely "novel" in an intellectually stimulating way, but still highly valuable that can be explored orders of magnitudes faster than is currently the case.
Then again, gassing rats and taking biopsies is not something you can do with AI.
Also, like, let’s maybe _not_ make the “gassing and cutting living organisms open” AI? Let’s just leave that particular genie in its bottle?
I'd bet you could 10x the number and still be in low single digit percentages of the US workforce. And it seems pretty likely that AI-enabled startups will also employ less people per-startup.
If AI causes a white-collar jobs apocalypse, I don't think startups are picking up the slack, although it'll plausibly cushion the blow somewhat for top-performing tech workers.
Ambitious goals and new discoveries happen via novelty-based search. Progress in scientific discovery is measured by how different/interesting the outcomes are, not by closeness to a predetermined goal.
Discovery is a creative search that preserves optionality, whereas optimization restricts optionality. In other words, you usually don't discover anything novel unless you're trying new things that don't appear connected to the goal in the first place. Would an ML optimization loop have discovered transformers?
1. There’s some irreducible costs in terms of time and material in the physical world that are not amenable to the kind of optimization or parallelization or even just the raw speedup from Moore’s law or computational architecture improvements we’re used to with software. My experience is primarily in biology, where the examples here are things like “it takes 20 minutes for E. coli to replicate” - it has taken 20 minutes for E. coli to replicate for a billion years, and next year it will still take E. coli 20 minutes to replicate, no matter how good your software stack is. Similarly, it takes X amount of energy to grow enough E. coli to produce a meaningful result, and that energy costs money, whether it’s in the form of glycerine or heat or whatever you want, and that also won’t materially reduce in the same kinds of “orders of magnitude” sense we’re used to from software, which is what we’re usually expecting to make the economics of these things work out.
2. Complicating the above, physical systems are phenomenally multivariate - far, far more than you think, and biological systems especially are just unbelievably complex - which means the number of experiments and the length and duration of those experiments you need to run to get enough data to be reasonably confident you’re seeing genuine signal is Way higher than you think.
Combine those two things and what you get is a money furnace, even before you get to the AI model training part, which is Also a money furnace. There’s low hanging fruits in all this, there’s areas where automating the approach can be really valuable, but typically the moment you turn this machine on, you’re gonna start burning money at a rate that would embarrass a finance bro on a coke bender, and that’s effectively unavoidable because the real world is not amenable to software’s scaling laws.