CEO & Co-founder, Synacti & Longshot Systems. Into Open Source, Software 2.0 and Statistics.
London
Joined January 2012
- Tweets271
- Following987
- Followers185
- Likes754
Mark Goodhead retweeted
I’m calling it. Jailbreaks will be solved in 2024.
RepEng LFG.
No LLM is secure! A year ago, we unveiled the first of many automated jailbreak capable of cracking all major LLMs. 🚨
But there is hope?!
We introduce Short Circuiting: the first alignment technique that is adversarially robust. 🧵
📄 Paper: arxiv.org/abs/2406.04313
Mark Goodhead retweeted
If you'd asked me a year ago, superposition would have been by far the reason I was most worried that mechanistic interpretability would hit a dead end.
I'm now very optimistic. I'd go as far as saying it's now primarily an engineering problem -- hard, but less fundamental risk.
Mark Goodhead retweeted
1/ 11 weeks ago I agreed to Chair the UK's efforts to accelerate state capacity in AI Safety - measuring and mitigating the risks of frontier models so we can safety capture their opportunities. Here is our first progress report: gov.uk/government/publicatio…
📢An elite team of high-profile AI specialists have been recruited as advisors for the Frontier AI Taskforce.
The Taskforce will protect against risks, build on UK capabilities & improve public services.
@MichelleDonelan met Ian Hogarth @soundboy, Taskforce Chair, to discuss
Mark Goodhead retweeted
Centuries after probability theory was developed, many philosophers still treat belief and justification as binary rather than continuous properties. Now the theory of logical induction allows us to treat rationality as continuous too, but who knows how long until that sinks in?
Mark Goodhead retweeted
You don't need to be deeply pessimistic about AI to think it's good to work on AI safety, just like you don't need to be deeply pessimistic about planes to think it's good to work on aviation safety. Risk reduction is valuable across a wide range of risk estimates.
Mark Goodhead retweeted
At OpenAI a lot of our work aims to align language models like ChatGPT with human preferences. But this could become much harder once models can act coherently over long timeframes and exploit human fallibility to get more reward.
📜Paper: arxiv.org/abs/2209.00626
🧵Thread:
Mark Goodhead retweeted
We're searching for a million people who love the NHS and think it's worth celebrating - pls RT and follow if you do #OurNHS
Mark Goodhead retweeted
Longshot Systems - Senior Machine Learning Researcher for fast growing startup (London, UK) kaggle.com/jobs/17352/longsh…
Mark Goodhead retweeted
How to tell if your dog is involved in a sex scandal.
thepoke.co.uk/2013/10/01/how…
Mark Goodhead retweeted
Saturday. A man in Shoreditch screams into a phone, frustrated by his inability to monetise flying ants.
Mark Goodhead retweeted
If your job title on Linkedin is 'Strategic Innovationiser' then I'm sorry but we can't be friends
Mark Goodhead retweeted
Aaron dead.
World wanderers, we have lost a wise elder.
Hackers for right, we are one down.
Parents all, we have lost a child. Let us weep.
Mark Goodhead retweeted
All time favourite front page headline in the Sunday Sport: GORDON RAMSAY SEX DWARF EATEN BY BADGER
Mark Goodhead retweeted
Billion-Ton Comet May Have Missed Earth by a Few Hundred Kilometers in 1883 j.mp/VGnKtN
Mark Goodhead retweeted
Whoever coined “there’s so such thing as a free lunch” evidently hasn’t being using the AmEx #foursquare promotion
Mark Goodhead retweeted
We've drastically improved the speed of Road travel searches. Only English roads for now, but we think you'll like it. locatable.com
Mark Goodhead retweeted
Just updated the tutorial to make things even easier - give it a twirl and let us know what you think: locatable.com