In recent months, the CEOs of leading AI companies have grown increasingly confident about rapid progress:
* OpenAI's Sam Altman: Shifted from saying in November "the rate of progress continues" to declaring in January "we are now confident we know how to build AGI"
* Anthropic's Dario Amodei: Stated in January "I'm more confident than I've ever been that we're close to powerful capabilities... in the next 2-3 years"
* Google DeepMind's Demis Hassabis: Changed from "as soon as 10 years" in autumn to "probably three to five years away" by January.
What explains the shift? Is it just hype? Or could we really have Artificial General Intelligence (AGI)[1] by 2028?
In this article, I look at what's driven recent progress, estimate how far those drivers can continue, and explain why they're likely to continue for at least four more years.
In particular, while in 2024 progress in LLM chatbots seemed to slow, a new approach started to work: teaching the models to reason using reinforcement learning.
In just a year, this let them surpass human PhDs at answering difficult scientific reasoning questions, and achieve expert-level performance on one-hour coding tasks.
We don't know how capable AGI will become, but extrapolating the recent rate of progress suggests that, by 2028, we could reach AI models with beyond-human reasoning abilities, expert-level knowledge in every domain, and that can autonomously complete multi-week projects, and progress would likely continue from there.
On this set of software engineering & computer use tasks, in 2020 AI was only able to do tasks that would typically take a human expert a couple of seconds. By 2024, that had risen to almost an hour. If the trend continues, by 2028 it'll reach several weeks.
No longer mere chatbots, these 'agent' models might soon satisfy many people's definitions of AGI — roughly, AI systems that match human performance at most knowledge work (see definition in footnote).
This means that, while the compa
The Meta-Research Innovation Center at Stanford (METRICS) is hiring post-docs for 2016/2017. The full announcement is available at http://metrics.stanford.edu/education/postdoctoral-fellowships. Feel free to contact me with any questions; I am currently a post-doc in this position.
METRICS is a research center within Stanford Medical School. It was set up to study the conditions under which the scientific process can be expected to generate accurate beliefs, for instance about the validity of evidence for the effect of interventions.
METRICS was founded by Stanford Professors Steve Goodman and John Ioannidis in 2014, after Givewell connected them with the Laura and John Arnold Foundation, who provided the initial funding. See http://blog.givewell.org/2014/04/23/meta-research-innovation-centre-at-stanford-metrics/ for more details.
(I tried posting this in a separate article, but as a new user I don't have enough karma. For now it is going to the open thread; if people think this should get more visibility I'd be happy to move it once I get sufficient karma)
That sounds really cool and should be shared in the EA job postings group
The talks from EA Global are available in podcast format at EARadio (iTunes link). The YouTube videos seem to have disappeared, so this may be the easiest way to access the talks. Cheers!
Here's an interesting question from Eva Vivalt of AidGrade on the FB group: "What are some important research questions that could use more work? Let's make a list!"
Some social change questions:
1) Is it better to focus on the "influencers," often through personalized contact and meetings?
2) Which countries are most important in steering global conversation?
3) Does controversy tend to reduce the growth ceiling of a movement by creating opposition (even though it, arguably, speeds up growth)?
Some meta stuff:
1) How can we spread EA effectively?
2) What are the lowest-hanging fruit for improving the EA movement?