Interesting! It also fails with a simple “conjugate ყოფნა” and fails with another Georgian verb, “conjugate გაკეთება,” meaning, to do, to make. Again, with all safety settings turned off.
RESEARCH
-
Ko-LLM Leaderboard Accepted at ACL 2024 Main Track
By
–
Great news! @upstageai Open Ko-LLM Leaderboard work is accepted at ACL 2024 Main track, thanks to @huggingface and @clefourrier .
-
Benchmark Smaug on LTM for Context Integration
By
–
You should benchmark Smaug on our LTM Benchmark, that focuses on integration of information in context
-
Building Worthy AI: Policy Steps, End State Vision, and Human Role
By
–
I would ask about: – specific steps that he thinks need to be taken (global politics, regulations, incentives) and what needs to be avoided – describe the end state in concrete terms (what is worthy successor and what is the role of humans there – why should we build worthy
-
New Institute IRAID Founded with Billion Dollar Funding
By
–
I’ve just raised a billion dollars from a couple of tech billionaires to fund a new institute, to be called IRAID – Institute for Raising the Alarm about Imaginary Dangers.
-
Private Message Training Lacks Intelligence and Raises Ethical Concerns
By
–
Training on private messages isn't accelerating though, it's just dumb
-
Data Decontamination and Generalization in Advanced Math Benchmarks
By
–
That's a valid point. The team tried hard to decontaminate the data. Also note that there's generalization into more challenging benchmarks such as HiddenMath and IMO-Bench.
-
OpenAI Explains Strategy Following Key Executive Departure
By
–
We’re really grateful to Jan for everything he's done for OpenAI, and we know he'll continue to contribute to the mission from outside. In light of the questions his departure has raised, we wanted to explain a bit about how we think about our overall strategy. First, we have
-
Gemini 1.5 Pro Achieves 90% MATH Benchmark Milestone
By
–
Also astonishing to see how fast the field has advanced: Gemini 1.5 Pro was first to surpass the 90% mark on MATH (6.9% 3 years ago). For comparison, on ImageNet, it took us close to 10 years to achieve the same from AlexNet (40%) to Meta Pseudo Labels (90.2%, our work
