My article on how (false) narratives on AI systems (dangerously) shape policy and legal discourse is now available *open-access* on Legal Studies' official page: https://t.co/sHF6HRIkSd. Quick explainer thread:
Prompt injection keeps getting wilder. Unlike text-based methods, you can't tell by looking at one of these images or audio clips that it's malicious. It can be embedded in a website/email attachment and will alter the model's behavior if it processes it.
https://t.co/NbzgKyZFv6
Back in 2021 @mattwadd wrote the below blog post and @mengwong took an example from that post and wrote it in L4. I’m now sitting with Matthew while he learns L4 using that same example while Meng instructs from the front at #CLAWCON
Loop closed!
Junior ML dev: Interesting problem. We need 3 months to gather user emails and link them to "hasattachments: yes" in our DB and train a model - should be a simple task, perhaps 3.5-turbo is enough
Senior ML dev: if "attach" in message: warn
If a robot that acts sufficiently like a human should be treated like a human, should a human that acts sufficiently like a robot be treated like a robot?
Brilliant Labs, a Singapore-based startup that develops augmented-reality wearables powered by ChatGPT, raised more than $3 million in a seed round from investors including cofounders of some of Silicon Valley’s best-known companies. https://t.co/TR3LMIUG9m
The rime of the average data scientist:
Training models overnight
Results saved for first light
Now imagine my utmost fright
"IOError: permission denied"
@yoavgo@CatalinaGoanta Which is yet different from "with the right prompt we can get the LLM to answer correctly some questions that involve X. Sometimes. If the time of day is right. And some tech company hasn't updated its model since we last tried. Or retrained it on data that now includes our test"
@aarontay What?! How would journals in those disciplines gatekeep away those who don't have the resources to get someone else to do the citations for them then?
BREAKING: The European Parliament has adopted the #AI Act with an overwhelming majority. No last-minute amendment outside the text adopted at the committee level made it, including those from the European People's Party on Remote Biometric Identification.
I wrote about the liberal & libertarian roots of Silicon Valley’s hostility towards democracy. The private control of technological innovation—steered towards surveillance, exploitation, and social control—is a existential risk, not AI. https://t.co/eVUj7Gl1SP