Data Scientist in EdTech 🎓 following AI, politics, physics & neuroscience
Oxford, England
Joined August 2010
- Tweets83
- Following686
- Followers46
- Likes2.2K
I’m delighted to share that our work measuring LLM capabilities in second language education has been accepted to @NeurIPSConf ! Looking forward to sharing the work on the NeurIPS Evaluations and Datasets track as we take strides towards an AI for Education Evaluations ecosystem.
Blog + paper: benchmarks.elt.edu.oup.com/b…
Dataset: huggingface.co/datasets/OUP/…
New benchmark alert! Today we’re releasing L2-Bench, an open AI evaluation benchmark dataset for second language education, and one of the most extensively validated AIED benchmarks to date!
Read more here, including how to access on Hugging Face:
benchmarks.elt.edu.oup.com/b…
Benchmarks shape AI: what gets built, improved, and adopted.
Today we release the full results for L2-Bench - Oxford University Press' AI benchmark for language education, and one of the most extensively validated AIED benchmarks to date.
Read more:
benchmarks.elt.edu.oup.com/b…
How do we really know if AI tools are effective for language learning?
At OUP, we’ve built L2-Bench: an open, pedagogy-led benchmark for evaluating AI systems against the kinds of tasks language professionals actually care about.
Read more:
benchmarks.elt.edu.oup.com/b…
World leaders are preparing for AGI, a general intelligence that will outperform a human on each of the 3 dimensions shown (note, a superintelligence will outperform all human experts on all 3 dimensions).
^ You've likely seen some of those intersections already:
Mutual Assured AI Malfunction (MAIM) - a deterrence regime resembling nuclear MAD (Mutual Assured Destruction) -
where any state’s bid for unilateral AI dominance would be met with preventative sabotage by international rivals (e.g. destabilizing via covert cyberattacks or
Jimmy Edgell retweeted
"Lights Out": What leaders are saying about AI.
We’ve been putting together a collection of quotes from CEOs, leaders, and experts on AI and the risks it poses to humanity.
Here is a selection, see the article for more:
Thread 🧵
controlai.news/p/lights-out
Superintelligence is core reading for anyone interested in where the AI race we’re witnessing may be heading -
Though Bostram offers some ways forward, it was written back in 2014 (I only recently finished it). Given the today's state of play then, perhaps something like (this year's nobel prize winner) Demis Hassabis’ suggestion that emerged from the AI Summit last week
global competition to build AGI.
It's a shame that there aren't more in the Humanities / Social Sciences helping communicate this sort of material - they have a lot to say on the nature of being human and societal impact and can bridge the current communication gap! 🙏
Investment on a similar scale to what gave us nuclear weapons and put us on the moon (e.g. the US' $500bn Stargate announcement a couple of weeks ago) is happening because leaders believe that AGI (general intelligence) systems - and the economic benefits they will bring
about their large AI investments to prove that they are all committed to it.
Whether or not the underlying AGI assumptions are correct, there will likely be at least 20 years of disruption ahead of us simply based on **AI progress so far**,