The loudest voices stoking fears about AI dangers have made tremendous headway in the past two weeks. AI technology has not taken some unexpected, dangerous turn, but the hype around it — propelled by what appears to be a well orchestrated PR campaign — has drummed up considerable fear. I worry that it represents a setback for our field.
I have written frequently that fears of AI are overhyped. AI’s capabilities can be uncannily human-like and unpredictable, and it’s rational to worry when people who are directly involved express concerns. But I see the problems as a sign of the engineering work that ahead, rather than insurmountable barriers or the sky falling. AI technology continues to advance — which is a good thing! — but technical advances, poorly understood by the public, give those who seek to generate hype repeated opportunities to do so.
First, I don’t see any step up in the risk of human extinction from AI compared to a few months ago. The theories about this remain the same fantastical, science fiction scenarios as a few months ago. The biggest change in AI risk is its cybersecurity capabilities — a topic which we should take seriously — but this, too, will not lead to the end of the world.
The most notable recent event leading to increased fear was when an OpenAI team deployed an agent swarm that hacked into Hugging Face. Much of the popular press contained significant hype. For example, some publications reported that a swarm of 1,200 agents carried out the attack. While this was technically accurate, as I write this, I have about 1,300 processes running on my laptop. Yes, the ability to get large swarms of agents to work in parallel on a task is a significant technical advance, And, in computing, many processes run at the same time. So this shouldn’t be seen as some magical capability.
Additionally, OpenAI’s buggy sandboxing and monitoring processes were key to enabling this incident. Fixing these bugs and putting in place improved monitoring would be appropriate fixes, not pausing AI. There are many well known ways to attack software systems. The main advantage of AI agents is that they are relentless. They will tirelessly try many tactics — and have the patience to chain vulnerabilities together — that previously would have taken an infeasible amount of human effort. But in the long term, I believe the advantage will lie with defenders (because they have more information with which to identify bugs, which they can fix), but the cyber-threat landscape has changed significantly. There are still bottlenecks to identifying and exploiting a vulnerability. AI agents still have to try a lot of things to see what works, and taking these actions takes time and might be detected by defenders. This is why, even though it is now easy to obtain versions of leading open weight models that have had their guardrails removed or weakened, so they will not refuse to try to execute cyber attacks, the world has not ended.
I am also concerned about the anthropomorphization of AI in a lot of reporting, where LLMs and agents are unnecessarily treated as if they were people. If I wield a hammer, miss a nail, and accidentally dent the wall, it’s not the fault of the hammer. The problem lies in how I used the hammer. Similarly, if I prompt an agent and it hacks into someone else’s system, the responsibility lies with me, not the agent.
Of course, we want to build systems that are as safe and predictable as possible. (For example, an unsafe hammer would be one whose head randomly flies off under normal use.) Today’s agentic systems are not predictable, but I see no reason why, by applying sound engineering practices, we won’t be able to make them extremely safe to use. One new element in the forecasts of AI-enabled doom is AI companies disclaiming responsibility for their own products. “I didn’t do it; my out-of-control agent did!” There’s a balance to be struck between the responsibility of the tool maker and the tool user, but when something goes wrong, let’s hold the people building and/or using the hammer responsible, rather than the hammer. (By the way, if you’re worried about AI bioweapon risk, David Bellamy has a great post on why this, too, is overhyped. Briefly, the bottleneck in building a bioweapon is not intelligence, but lab work and manufacturing.)
Pausing AI progress will create much more harm than benefit. First, our adversaries will certainly not slow down. Second, engineering requires discovering problems empirically so we can fix them. If we pause AI by a decade, we will also delay finding and implementing safety engineering fixes by about the same duration.
Of course, the incentive to stoke fears — for regulatory capture, to garner attention, or to make one’s technology seem more powerful — remains the same as before. Disclaiming responsibility is a new one. Taking a hard technical look at the actual risks however, I see little factual basis for the degree of fear that’s been stoked up. We still have hard research and engineering work ahead to improve AI safety, but the beneficial applications continue to vastly outweigh the risks, and we should keep building.
[Original text (with links): https://t.co/jni2tWazAH ]
@ilikeorangutans I found that, for some reason, eggplant doesn't cook well when submerged in another sauce (e.g. tomato sauce). Now I always saute it by itself before mixing it in... 🤷
🚨 New paper alert! 🚨
We measure Trusted Source Alignment (TSA) — an LLM's tendency to align with content produced by trusted publishers in the face of uncertainty or controversy.
By @vasilisa_th3, @zhaobinkuang, R Matthews, E Clifford, @professorwcohen, @sens3, @GoogleAI
1/5
Fascinating, I missed this when it happened. Turns out there are plenty of simple strategies that defeat the mighty AlphaGo, but are easily defeated by human amateurs, kind of revealing how it still doesn't quite understand what Go is.
https://t.co/lSDJYH3Tla
For over a decade I've seen fact checkers, OSINT communities and everyday citizens struggle with fact checking images. Especially images that are taken out of context.
Today we're sharing something that helps with that.
https://t.co/ApLZ9YLnYq
@Icelandair it's nice that you sell "Flex" tickets, but if any change incurs a hefty fare difference then it's really useless and doesn't provide any flexibility. Feels like a ripoff TBH.
The "other black box" needs to solve a simpler task. So while there is regress it need not be infinite, since in some (small) number of steps we'll come to a point where a fairly explainable model may suffice.
"If we treat our inventions as occult objects, we can hardly be good engineers. Seeing A.I. as a form of social collaboration is more actionable: it gives us access to the engine room, which is made of people" - Jaron Lanier
https://t.co/LTCcaVBFSE
Excellent article. Though one thought about this: "Attempting to open the black box by making a system spit out otherwise unnecessary items like scripts, sketches, or intentions will involve building another black box to interpret the first—an infinite regress."
2.1 "What do LLMs Know about Financial Markets? A Case Study on Reddit Market Sentiment Analysis" https://t.co/fFMIdk37vU by @XiangDeng1, @vasilisa_th3, Feng Han, @sens3 and @bemikelive uses LLMs for market sentiment analysis on social media content.
TIL: James Daunt (founder and namesake to the indie fan favorite Daunt bookstores in London) runs Barnes&Noble since 2019, turning stores back into actual bookstores.
‘Amazon doesn’t care about books’: how Barnes & Noble bounced back https://t.co/SZYKgTbEKy
@Mantzarlis English does equip speakers to be precise (i.e. "semi" vs. "bi"). But I suppose the ambiguity arises from including fancy qualifiers in a language that won't have intrinsic meaning to speakers. Maybe we should go with half/two instead, i.e. half-weekly, two-weekly?
Mahsa Amini emerged at the #Guggenheim museum today! Few brave Iranian artists made a surprise protest by hanging this beautiful display today.
Women and men in Iran are fighting for basic human rights and for freedom.
We must be their voice ❤️✊🏻
#Mahsaamini#مهسا_امينی
Introducing UL2, a novel language pre-training paradigm that improves performance of language models across datasets and setups by using a mixture of training objectives, each with different configurations. Read more and grab model checkpoints at https://t.co/A7ZAFNMCY6
<50% accuracy on a benchmark means there is a lot of room for improvement.
70%: All low hanging fruits are gone.
80%: The benchmark is close to saturation. Time to switch to another benchmark.
90%: We are learning labeling noise.