Axiom Futures AI Safety Week 2
Why are people building AI systems article
Automating tasks
McKinsey estimate of 0.1 to 0.6% growth rate in productivity (!?) which leads to a value addition of $2.6 to 4.4 Trillion dollars.
- This is a disputed figure. Original McKinsey article estimates use cases such as hiring which people are quite skeptical about because of problems with fairness or hallucinations.
- Earlier AI systems were thought to have good use cases in operations and supply chain but those were more compute/calculation based optimizations which are now not considered as relevant because of generative AI.
- Customer Operations, Sales, Software Engineering, R&D. (only SWE and R&D seem legit).
Replacing Human workers
e.g. Devin or the idea of a “drop in” agent as mentioned by Leopold Aschenbrenner
Non Economic Motivations
- Positive externalities of technology development
- National Interests (proprietary models and when AGI comes its going to be like nuclear power).
- Fun Stuff / Solving their own problems.
Compute Trends across three eras of ML article
Better ML systems come from:
- Better Algorithms.
- Better Data.
- More Compute.
This article aims to analyse training compute trends.
Research Methodology
How is compute trained:
- Counting number of operations.
- Estimating through Runtime - requires assumptions about GPUs and system used for training - more uncertain.
Only the “important models” (influence, number of citations) were considered that pushed the SoTA.
Results
- Old era doubling time ~20 months.
- DL era (~2010-12) doubling time of around 6 months.
- Large Model Era (~2016) doubling time of around 10 months.
Scaling Laws article
Compute, Data, Parameters. Neural Language models scale with a consistent power law wrt to these. Transformer architecture doesn’t converge.
Paper on AI prediction
High-level machine intelligence (HLMI) is achieved when unaided machines can accomplish every task better and more cheaply than human workers.
3.6 Intelligence Explosion
Since 2016 a majority of respondents have thought that it’s either “quite likely,” “likely,” or an “about even chance” that technological progress becomes more than an order of magnitude faster within 5 years of HLMI being achieved.
3.7 Future Systems
There were areas of agreement:
- Find unexpected ways to achieve goals (82.3% of respondents)
- Be able to talk like a human expert on most topics (81.4%)
- Frequently behave in ways that are surprising to humans (69.1%)
Interpretability / Explainable AI systems
Only ~20 percent think that it’s likely that for typical state-of-the-art AI systems in 2028, it will be possible for users to know the true reasons for systems making a particular choice.
Prediction Exercise
My Predictions:
- Event: Win Putnam Math Contest. Prediction: 75% confidence by 2027. Reasoning: We have already done pretty decent on IMO benchmarks.
- Event: Simple Python Code given spec. Prediction: 100% confidence 2023. Reasoning: Claude 3 Opus and GPT-4 pretty much solve this.
- Event: Finetuning an LLM. Prediction: 75% confidence 2028. Reasoning: With good prompting it is almost able to do so.
Partner Predictions:
- Vid from diff angle. Prediction: 75% confidence by 2028. Reasoning: 2020 panoptic scene graph, SoRA.
How will AI affect your job Exercise
- Reading research papers or textbooks, coding, solving physics problems, interacting with others.
- In 10 years: Better comprehension, significantly fewer bugs in coding, good progress in research-level problems, significant AI-generated text/speech.
- Technology adoption would be quite slow. Even with all the developments there would be a significant transition delay.
- AI would exacerbate the rockstar effect considerably but also empower creatives.
Question for the week: Who is liable for the damages if an AI system generates something that wouldn’t have been possible otherwise?