I predict that by the end of 2025 neural nets will:
- have human-level situational awareness (understand that they're NNs, how their actions interface with the world, etc)
- beat any human at writing down effective multi-step real-world plans
- do better than most peer reviewers
Conversation
- autonomously design, code and distribute whole apps (but not the most complex ones)
- beat any human on any computer task a typical white-collar worker can do in 10 minutes
- write award-winning short stories and publishable 50k-word books
- generate coherent 20-min films
11
23
301
The best humans will still be better (tho much slower) at:
- writing novels
- robustly pursuing a plan over multiple days
- generating scientific breakthroughs, including novel theorems (tho NNs will have proved at least 1)
- typical manual labor tasks (vs NNs controlling robots)
13
17
260
FWIW my actual predictions are mostly more like 2 years, but others will apply different evaluation standards, so 2.75 seems more robust.
Also, they're not based on any OpenAI-specific information.
For my definition of situational awareness see
2
8
174
Lots to disagree with here ofc. I'd be particularly interested in:
A) people giving median dates they expect these to be achieved
B) people generating other specific predictions about what NNs will and won't be able to do in a few years' time
9
2
111
Re the planning one, since a lot of people are commenting on that:
Quote Tweet
Replying to @MugaSofer and @anthrupad
Two clarifications here:
- I think writing down plans doesn't get you very far, the best plans are often things like "try X, see what happens, iterate".
- It's about beating any human (across many domains) not beating the best human in each domain.
3
2
72
And another clarification on planning:
Quote Tweet
Replying to @RichardMCNgo @MugaSofer and @anthrupad
By "many domains" I don't mean literally all of them, but a pretty wide range. E.g. averaged across all businesses that McKinsey has been hired to consult for, AI will make better business plans than any individual human could.
See already: twitter.com/emollick/statu
2
2
50
One terminology clarification: when I say "I predict" I mean I assign >50% credence, but not necessarily *much* more than 50%.
That's inherently vague but so are my opinions, so I'm not sure there's much benefit from trying to be more concrete on probabilities and timelines.
8
4
72
One more prediction: NNs will pass the current version of the ARC autonomous replication evals (see section 2.9 of the GPT-4 system card; page 55: arxiv.org/abs/2303.08774). But they won't be able to self-exfiltrate from secure servers, or avoid detection if cloud providers try.
1
3
41
Another prediction: 5% of adult Americans will report having had multiple romantic/sexual interactions with a chat AI, and 1% having had a strong emotional attachment to one.
9
1
38
We'll see clear examples of emergent cooperation: AIs given a complex task (e.g. write a 1000-line function) in a shared environment cooperate without any multi-agent training.
Also: after a chat with I've revised my video generation prediction to 5 not 20 mins.
2
26
But looks like I made the classic mistake of indexing only on what was already out, not on the stuff that was about to be released:
Quote Tweet
Pushed #gen2 again & made a movie trailer.
#aicinema is finally here!
Every shot made from text prompts, except one iconic shot you all know, done with #gen1
Made possible by @runwayml
@bazluhrmann your movies been a great inspiration! 

Voices: @elevenlabsio
#aianimation
Show this thread
1:30
7.4M views
1
17

