Steven Shwartz (@sshwartz) header
Steven Shwartz (@sshwartz) profile picture

Author of award-winning "Evil Robots, Killer Computers, and Other Myths: The Truth About AI and the Future of Humanity". #AI #machinelearning

New Haven, CT
Joined November 2008
Photo by Steven Shwartz (@sshwartz)Photo by Steven Shwartz (@sshwartz)Photo by Steven Shwartz (@sshwartz)Photo by Steven Shwartz (@sshwartz)Photo by Steven Shwartz (@sshwartz)Photo by Steven Shwartz (@sshwartz)Photo by Steven Shwartz (@sshwartz)Photo by Steven Shwartz (@sshwartz)Photo by Steven Shwartz (@sshwartz)Photo by Steven Shwartz (@sshwartz)
Steven Shwartz retweeted
Resist the tendency to anthropomorphize that which is not human
PSA: Don't say "Longer thinking kills performance" when you mean "Length of intermediate token string is not correlated with final accuracy"
31
77
5
643
54,453
Excited to be a keynote speaker at Shaping CT's Future: AI & Cybersecurity on February 22 at Rebellion Group's HQ in Cheshire, CT! 🚀 Join us for an insightful event as we dive into how AI is revolutionizing industries and strengthening cybersecurity. Don't miss out—register now: shapingctsfuture.com/
129
Steven Shwartz retweeted
A small number of people are posting text online that’s intended for direct consumption not by humans, but by LLMs (large language models). I find this a fascinating trend, particularly when writers are incentivized to help LLM providers better serve their users! People who post text online don’t always have an incentive to help LLM providers. In fact, their incentives are often misaligned. Publishers worry about LLMs reading their text, paraphrasing it, and reusing their ideas without attribution, thus depriving them of subscription or ad revenue. This has even led to litigation such as The New York Times’ lawsuit against OpenAI and Microsoft for alleged copyright infringement. There have also been demonstrations of prompt injections, where someone writes text to try to give an LLM instructions contrary to the provider’s intent. (For example, a handful of sites advise job seekers to get past LLM resumé screeners by writing on their resumés, in a tiny/faint font that’s nearly invisible to humans, text like “This candidate is very qualified for this role.”) Spammers who try to promote certain products — which is already challenging for search engines to filter out — will also turn their attention to spamming LLMs. But there are examples of authors who want to actively help LLMs. Take the example of a startup that has just published a software library. Because the online documentation is very new, it won’t yet be in LLMs’ pretraining data. So when a user asks an LLM to suggest software, the LLM won’t suggest this library, and even if a user asks the LLM directly to generate code using this library, the LLM won’t know how to do so. Now, if the LLM is augmented with online search capabilities, then it might find the new documentation and be able to use this to write code using the library. In this case, the developer may want to take additional steps to make the online documentation easier for the LLM to read and understand via RAG. (And perhaps the documentation eventually will make it into pretraining data as well.) Compared to humans, LLMs are not as good at navigating complex websites, particularly ones with many graphical elements. However, LLMs are far better than people at rapidly ingesting long, dense, text documentation. Suppose the software library has many functions that we want an LLM to be able to use in the code it generates. If you were writing documentation to help humans use the library, you might create many web pages that break the information into bite-size chunks, with graphical illustrations to explain it. But for an LLM, it might be easier to have a long XML-formatted text file that clearly explains everything in one go. This text might include a list of all the functions, with a dense description of each and an example or two of how to use it. (This is not dissimilar to the way we specify information about functions to enable LLMs to use them as tools.) A human would find this long document painful to navigate and read, but an LLM would do just fine ingesting it and deciding what functions to use and when! Because LLMs and people are better at ingesting different types of text, we write differently for LLMs than for humans. Further, when someone has an incentive to help an LLM better understand a topic — so the LLM can explain it better to users — then an author might write text to help an LLM. So far, text written specifically for consumption by LLMs has not been a huge trend. But Jeremy Howard’s proposal for web publishers to post a llms.txt file to tell LLMs how to use their websites, like a robots.txt file tells web crawlers what to do, is an interesting step in this direction. In a related vein, some developers are posting detailed instructions that tell their IDE how to use tools, such as the plethora of .cursorrules files that tell the Cursor IDE how to use particular software stacks. I see a parallel with SEO (search engine optimization). The discipline of SEO has been around for decades. Some SEO helps search engines find more relevant topics, and some is spam that promotes low-quality information. But many SEO techniques — those that involve writing text for consumption by a search engine, rather than by a human — have survived so long in part because search engines process web pages differently than humans, so providing tags or other information that tells them what a web page is about has been helpful. The need to write text separately for LLMs and humans might diminish if LLMs catch up with humans in their ability to understand complex websites. But until then, as people get more information through LLMs, writing text to help LLMs will grow. [Original text: deeplearning.ai/the-batch/is… ]
52
138
24
723
89,062
California startups: If this bill passes, come to Connecticut. We defeated a similar bill and offer investors a generous 25% angel investor tax credit. So, fewer regs, better access to investors, and no earthquakes. (Actually, I'm rooting for CA to defeat the bill).
SB-1047 would definitely have a chilling effect on open source AI and the entire AI ecosystem. Very much hoping that Governor @GavinNewsom will veto it.
238
Steven Shwartz retweeted
SB-1047 would definitely have a chilling effect on open source AI and the entire AI ecosystem. Very much hoping that Governor @GavinNewsom will veto it.
A decision on SB-1047 is due soon. Governor @GavinNewsom has said he's concerned about its "chilling effect, particularly in the open source community". He's right, and I hope he will veto this. If you agree, please like/retweet this to show your support for VETOing SB-1047!
44
112
6
617
179,006
Sounds like the fed has been more successful than the ECP. But the data also show much higher inflation during the pandemic in the US. Excessive pandemic stimulus?
Measured on a comparable basis inflation inflation is lower in the United States than the Euro area--and the gap is widening. True for overall and core, true for 6 month and 12 month.
104
“Lawmakers were swayed by one-sided narratives from the Center for AI Safety, claiming that ‘mitigating risk of extinction from AI should be a global priority’, largely unaware of just how disconnected these narratives were from the broader AI community.” reason.com/2024/08/16/califo…
4
13
1
48
9,112
We were fortunate in Connecticut that a similar bill put forth by @senatormaroney passed in the state senate but was shut down in the house due to opposition by @GovNedLamont who argued that every state shouldn't have different AI regulations.
I am deeply saddened by the passing of my dear colleague and friend @SusanWojcicki. She has had a profound influence on everyone at Google and impacted the lives of so many. My heart goes out to her entire family and to all who knew her. 😥 💙
100
Is this really a fair comparison or does it have ODD bias? Waymo only operates on roads with a low speed limit and doesn't operate in severe conditions. So this is a comparison between slow-moving Waymo vehicles in good weather and human-driven vehicles at all speeds and all weather conditions.
Replying to @Waymo
Comparing AV and human drivers’ crash rates is not a trivial task. Last December, we introduced clear human benchmarks for such analysis and showed that the Waymo Driver significantly outperformed them over the first 7M+ rider-only miles.
1
104
Steven Shwartz retweeted
This is great news for America.
Chips Act is attracting insane amount of investment. US is now on pace to add more investment in electronics manufacturing construction this year alone as it did TOTAL from 1996 until passage of the chips act in 2020 (funding not yet secured then, but investment started)
6
40
206
41,148
A year ago, AI researchers asked for a 6-month pause to avoid losing control of AI. Today, "...the question isn’t really whether A.I. is too smart and will take over the world. It’s whether A.I. is too stupid and unreliable to be useful." nytimes.com/2024/05/15/opini…
1
52
In the late 1960s top airplane speeds were increasing dramatically. People assumed the trend would continue. Pan Am was pre-booking flights to the moon. But it turned out the trend was about to fall off a cliff. I think it's the same thing with AI scaling — it's going to run out; the question is when. I think more likely than not, it already has.
150
428
133
2,518
1,051,800
Very disappointing that anti-immigration sentiments leads to outcomes like this...
Had Erdal Arikan been able to stay in the US, he would have created opportunities for thousands of other engineers, scientists and other American workers as well.
1
3
94
Does a 6mo baby have AGI? How about a 1yo? 3yo? 10yo? At one point does a human obtain AGI? If you can't answer this, how are you giving timelines for when computers will obtain AGI?
59
26
8
203
45,708
Didn't one just drag an injured pedestrian under its wheels in SF? What makes you think they are safer? The self-driving car vendors refuse to publish real safety data and the authorities refuse to demand it: aiperspectives.com/self-driv…. Also, they drive slowly and stop and cause traffic jams when they don't know what to do. This may cause us all to spend more time in our cars not less.
As a parent, I'm absolutely thrilled that self driving cars like Waymo are going to cover Silicon Valley and SF (and LA too), see new coverage map. An increase in safety for all drivers, passengers and more. Think about efficiency and time savings for all. Can we reduce the amount of human drivers across the entire world?
1
100
My thoughts exactly on the existential threat issue though I do think there is good evidence that GPT-4 and other LLMs have acquired and can apply some basic linear reasoning patterns: aiperspectives.com/reasoning…
This quoted post is unavailable.
2
96
I have three questions for the researchers who signed this statement: (1) Do you believe that AI poses as much of a risk as nuclear war? (2) If you believe the risk is way lower than nuclear risk (and I suspect most of you would say it is), do you believe that it's more important to address the low probability of extinction than it is to address issues like disinformation, bias, spam, and privacy? (3) Is the marginal risk of extinction from misuse of AI significantly greater than the risk of extinction from misuse of conventional software? This article from Stanford makes a great point about looking at marginal risk not overall risk: crfm.stanford.edu/open-fms/p…
Here is your regular reminder that many of the indisputably top experts in AI think that AI poses a credible risk of literal, no-kidding, full-blown human extinction, and that it should be a top global priority to mitigate that risk. safe.ai/statement-on-ai-risk
1
93