We're talking about graduate research programs in computer science. Also, I did all my post-secondary education in the US. I know the US system pretty well.
RESEARCH
-
Latest LLM Reads Text on Wall Warning of Impact
By
–
The latest LLM can read writing on a wall. It says "You're about to hit me."
-
AI Agents in Metaverse: CICERO Helping Humans Achieve Goals
By
–
Imagine an AI agent teaching you a new skill in the metaverse. Or a non-player character helping you find your way to a win in a video game. How ever developers responsibly build on CICERO’s code, there’s potential for AI to help humans reach more of their goals. #CICERObyMetaAI
— AI at Meta (@AIatMeta) 25 novembre 2022Imagine an AI agent teaching you a new skill in the metaverse. Or a non-player character helping you find your way to a win in a video game. How ever developers responsibly build on CICERO’s code, there’s potential for AI to help humans reach more of their goals. #CICERObyMetaAI
-
Research Mentorship for Undergraduates at Less Research-Active Schools
By
–
Right. But at that point, at a good school, hopefully a research mentor should be able to guide them. Still useful for undergrads at schools which are less research-active, or for students with absent research mentors.
-
Hallucination Risk in Generative AI Products and Market Response
By
–
Hallucination is an existential risk to any generative AI product, and people are blindly (irresponsibly?) forging ahead (and we're all tired and wary of "haha this was all AI!" rugpulls). Observing the different reaction to @MetaAI
's Galactica vs @metaphorsystems is instructive -
Generative AI framing undersells actual capabilities
By
–
Similar issue with "Generative AI" – so hot right now, but this framing is, believe it or not, *underselling* capabilities we are already seeing, and perhaps causing people to work on the wrong things.
-
Prompt Engineering Is a Kludge: Why It’s Just Temporary
By
–
#StableDiffusion2 is all the hotness right now, but there'll be new versions every few months. I wanted to zoom out and talk about the open secret that all AI insiders seem to know: Prompt engineering is just a kludge. Quotes from @gwern
, @sama
, @GaryMarcus
, @josephofiowa
: -
InstructGPT/RLHF tuning makes model assume all questions answerable
By
–
My guess is this is InstructGPT/RLHF rather than anything in the pre-training corpus. Tuning implicitly makes it assume all questions are answerable — it sees all text as “ ” and Q/A is a subset of that.
-
Fine-tuning Speech Recognition Models with Open Datasets
By
–
This is a challenge for sure. However, in some cases it is possible to find an academic/ open source dataset for the language that you’re interested in fine tuning. There is also always the possibility of using a Wav2Vec2.0 checkpoint to transcribe YT videos & use that.
