Share this page!

Enter URL or ID to Unroll

You can paste full URL like: https://x.com/threadreaderapp/status/1644127596119195649
or just the ID like: 1644127596119195649

How to get URL link on X (Twitter) App

  1. On the Twitter thread, click on or icon on the bottom
  2. Click again on or Share Via icon
  3. Click on Copy Link to Tweet
  4. Paste it above and click "Unroll Thread"!
  5. More info at Twitter Help
Haydn Belfield Profile picture
Haydn Belfield
@HaydnBelfield
Research Scientist (Frontier Planning) at @GoogleDeepMind. Research Affiliate @Cambridge_Uni @CSERCambridge & @LeverhulmeCFI. All views my own.
Mar 21, 2023 • 20 tweets • 6 min read
How big a deal is the new Alpaca language model? How freaked out should we be? I’ve been asked that a lot over the last few days. This thread 🧵is an initial stab at an answer. I’m still thinking through & would be grateful to hear others’ views! Alpaca fine-tuned Meta’s language model LLaMa to make it better at following instructions. They did by using GPT-3.5 to speed up the process.

Some people have got very worried by this step

Feb 1, 2023 • 4 tweets • 2 min read
Developing standards in EU & US over the next 2 years will be hard, techical work.
But this should be a call to arms - to get to work!

We can't yet say creating effective standards is "near-impossible": need to try first. Solution of ‘not using neural networks’ is infeasible The article spends a lot of time on problems but these all underline the *importance* of standards.
Doesn't need to be perfect in 2 years, just needs to be better than what we have now: nothing!

See eg this post for lots of good safety ideas
aligned.substack.com/p/alignment-op…
Nov 22, 2022 • 4 tweets • 4 min read
Having read the paper & supplementary materials, watched narrated game & spoken to one of the human players I'm pretty concerned.
The @ScienceMagazine paper centres 'human-AI cooperation' & the bot is not supposed to lie. However, videos clearly show deception/manipulation

1/3 Screenshots of the stab below.

The human player said:
"The bot is supposed to never lie [...] I doubt this was the case here"
"I was definitely caught more off guard as a result of this message; I knew the bot doesn't lie, so I thought the stab wouldn't happen."

2/3
Nov 15, 2021 • 13 tweets • 3 min read
1st paper with @shin10173 published!
‘AI & Antitrust: Reconciling Tensions between Competition Law & Cooperative AI Development’ (Yale Journal of Law & Tech).
We examine 14 forms of cooperation strategies, identify tensions & suggest mitigation steps
yjolt.org/ai-antitrust-r… 🧵 Cooperation between AI companies can help them create systems that are safe, secure & with broadly shared benefits. However the goal of comp law is to protect competition between rival companies. Potential conflicts are significant but currently underexplored. The 14 coop forms: