
Hannu Rajaniemi
@hannu
Followers
6K
Following
11K
Media
202
Statuses
7K
Co-founder and CEO of @HelixNano. Author of The Quantum Thief series, Summerland and DARKOME https://t.co/o6RsVIB7xb
San Francisco, CA
Joined January 2007
Our immune system is amazing, but faces pandemics, cancer and aging. It's also wired to reject what we need to extend our healthspans & minds: organs, implants, new genes. It's time to radically augment it. We need to build an immune-computer interface (ICI). A manifesto🧵!
23
153
810
I love the fact that the sentence "to prevent the model from becoming evil, we actually inject it with evil" here is a somewhat accurate description of complex AI alignment work.
We introduce a method called preventative steering, which involves steering towards a persona vector to prevent the model acquiring that trait. It's counterintuitive, but it’s analogous to a vaccine—to prevent the model from becoming evil, we actually inject it with evil.
0
1
23
If there are LLM vaccines, what are the LLM immune systems like?.
In which the gang (@RunjinChen, @andyarditi, @Jack_W_Lindsey ):. - identifies vectors for bad personas (evil, sycophancy, hallucinations, etc).- shows that if you inject the bad vectors in training, the model learns to not do the bad thing!!. aka vaccines but for LLMs.
1
1
7
RT @_Dave__White_: the openai IMO news hit me pretty heavy this weekend. i'm still in the acute phase of the impact, i think. i consider my….
0
349
0
RT @michael_nielsen: What project would you be happy to devote 50 years of your life to, despite (or because of) it only being 10% done whe….
0
20
0
Really interesting thread -- I feel this was all obviously coming when the OpenCRISPR paper came out, if not before.
Will AI-designed proteins break biologics patents?. If your IP strategy still relies on “≥80 % sequence identity,” it might be time for a rethink. Thread👇. Tools like ProteinMPNN from the Baker Lab, take an existing e.g. AbAg complex or enzyme and can replace substantial parts
3
1
15
Been thinking about this for a while! Interesting major CS experiments, at least on the AI side, might also now on average cost *more* than biology experiments of comparable impact.
It’s kind of interesting, but in some ways AI is making CS much more like biology. We can observe and tweak what it does, but we really don’t know how it’s doing it. A science and engineering discipline rooted in empiricism over theory. I wonder if this trend continues.
0
1
12
RT @fermatslibrary: Alan Turing was born 113 years ago today. • Cracked the German's Enigma code in WWII.• Proposed the “Turing Test” to j….
0
569
0