Seth Lazar
@sethlazar.org
Philosopher working on normative dimensions of computing and sociotechnical AI safety.
Lab: https://mintresearch.org
Self: https://sethlazar.org
Newsletter: https://philosophyofcomputing.substack.com
Lab: https://mintresearch.org
Self: https://sethlazar.org
Newsletter: https://philosophyofcomputing.substack.com
And there's a great paper by @_FelixSimon_ and Sacha Altay on the actual impact of generative AI on elections. And more that I'll be writing about later, (incl great art from Seb Krier).
Read them all at buff.ly/zTrRirO. Thanks so much to @knightcolumbia (and especially Katy) for making it happen
Read them all at buff.ly/zTrRirO. Thanks so much to @knightcolumbia (and especially Katy) for making it happen
September 5, 2025 at 4:21 PM
And there's a great paper by @_FelixSimon_ and Sacha Altay on the actual impact of generative AI on elections. And more that I'll be writing about later, (incl great art from Seb Krier).
Read them all at buff.ly/zTrRirO. Thanks so much to @knightcolumbia (and especially Katy) for making it happen
Read them all at buff.ly/zTrRirO. Thanks so much to @knightcolumbia (and especially Katy) for making it happen
My symposium on AI & Democratic Freedoms (edited with @katygb.bsky.social) is shaping up amazingly. @random_walker and @sayashk 's already influential 'AI as Normal Technology' is there. So is @danielsusskind's insightful investigation of what will remain for humans to do in our automated future.
September 5, 2025 at 4:21 PM
My symposium on AI & Democratic Freedoms (edited with @katygb.bsky.social) is shaping up amazingly. @random_walker and @sayashk 's already influential 'AI as Normal Technology' is there. So is @danielsusskind's insightful investigation of what will remain for humans to do in our automated future.
Building democratic resilience for the era of AI agents, and for AGI beyond, is an urgent challenge. If you're building civic agents, I'd love to talk.
The paper—written with the visionary Mariano-Florentino Cuéllar—is published here: buff.ly/dMM0r7K
The paper—written with the visionary Mariano-Florentino Cuéllar—is published here: buff.ly/dMM0r7K
September 5, 2025 at 4:21 PM
Building democratic resilience for the era of AI agents, and for AGI beyond, is an urgent challenge. If you're building civic agents, I'd love to talk.
The paper—written with the visionary Mariano-Florentino Cuéllar—is published here: buff.ly/dMM0r7K
The paper—written with the visionary Mariano-Florentino Cuéllar—is published here: buff.ly/dMM0r7K
More important still, if we want to preserve democratic values in this radical period of transition, is to make democratic institutions more resilient to the changes ahead. This can't just be about going back to how things were. Their stressors are not all exogenous.
September 5, 2025 at 4:21 PM
More important still, if we want to preserve democratic values in this radical period of transition, is to make democratic institutions more resilient to the changes ahead. This can't just be about going back to how things were. Their stressors are not all exogenous.
But computing has always been janus-faced for democracy, and this time won't be different. We could build civic agents that advance democratic values and disrupt concentrated power. Some features of modern AI could help. But civic agents must be built, they're not the default.
September 5, 2025 at 4:21 PM
But computing has always been janus-faced for democracy, and this time won't be different. We could build civic agents that advance democratic values and disrupt concentrated power. Some features of modern AI could help. But civic agents must be built, they're not the default.
And there will be novel threats—the erosion of cognitive autonomy, accelerated cyberwarfare, the ability for executives to wear the administrative state like a mech suit that implements without question their every whim.
September 5, 2025 at 4:21 PM
And there will be novel threats—the erosion of cognitive autonomy, accelerated cyberwarfare, the ability for executives to wear the administrative state like a mech suit that implements without question their every whim.
Soaring inequality? Check.
Concentration of corporate power? Also check.
Stuffed-up information ecosystem? Yep that too.
Backsliding as the 'autocratic legalism' playbook gets rolled out in one nation after the next? Agents could be a helpful software Stasi.
Concentration of corporate power? Also check.
Stuffed-up information ecosystem? Yep that too.
Backsliding as the 'autocratic legalism' playbook gets rolled out in one nation after the next? Agents could be a helpful software Stasi.
September 5, 2025 at 4:21 PM
Soaring inequality? Check.
Concentration of corporate power? Also check.
Stuffed-up information ecosystem? Yep that too.
Backsliding as the 'autocratic legalism' playbook gets rolled out in one nation after the next? Agents could be a helpful software Stasi.
Concentration of corporate power? Also check.
Stuffed-up information ecosystem? Yep that too.
Backsliding as the 'autocratic legalism' playbook gets rolled out in one nation after the next? Agents could be a helpful software Stasi.
Capable LLM-based agents are already here. For any domain where you can build a good RL verifier, current knowledge will get us to human-level performance and better. Further advances are on the horizon. Agents are bound to exacerbate the trends already stressing democracies.
September 5, 2025 at 4:21 PM
Capable LLM-based agents are already here. For any domain where you can build a good RL verifier, current knowledge will get us to human-level performance and better. Further advances are on the horizon. Agents are bound to exacerbate the trends already stressing democracies.
How will AI agents impact democratic values? Democracies are—for independent reasons—already under acute pressure. Since WWII Moore's Law and democratisation went up and to the right in lockstep. Not any more.
September 5, 2025 at 4:21 PM
How will AI agents impact democratic values? Democracies are—for independent reasons—already under acute pressure. Since WWII Moore's Law and democratisation went up and to the right in lockstep. Not any more.
Busy shopping day in Causeway Bay (long exposures handheld with Spectre App)
December 23, 2024 at 3:04 AM
Busy shopping day in Causeway Bay (long exposures handheld with Spectre App)
Really love the container ships we can see from our balcony. We now have an app to track them…
December 17, 2024 at 9:19 AM
Really love the container ships we can see from our balcony. We now have an app to track them…
Much enjoyed first @neuripsconf.bsky.social—fantastic to see the number (in absolute if not relative terms) of ML researchers seriously concerned with questions of ethics and safety. The coming months and years will try your integrity: I’m hopeful you’ll hold fast!
December 17, 2024 at 12:55 AM
Much enjoyed first @neuripsconf.bsky.social—fantastic to see the number (in absolute if not relative terms) of ML researchers seriously concerned with questions of ethics and safety. The coming months and years will try your integrity: I’m hopeful you’ll hold fast!
This is cool but there’s a logical leap here from ‘it’s complicated’ to ‘it’s all subjective’. I think what you’re probably after is just a kind of reasonable pluralism, rather than ‘anything goes’, right?
November 30, 2024 at 9:48 PM
This is cool but there’s a logical leap here from ‘it’s complicated’ to ‘it’s all subjective’. I think what you’re probably after is just a kind of reasonable pluralism, rather than ‘anything goes’, right?
Sold the trailer we did our Big Lap in today, to a young family planning to head off for nine months next year. Such a wonderful time we had, and I love that this is a rite of passage here in 🇦🇺
November 27, 2024 at 10:33 AM
Sold the trailer we did our Big Lap in today, to a young family planning to head off for nine months next year. Such a wonderful time we had, and I love that this is a rite of passage here in 🇦🇺
This paper by Yu Gu et al looks *fascinating* arxiv.org/abs/2411.06559
“Language agents have demonstrated promising capabilities in automating web-based tasks, though their current reactive approaches still underperform largely compared to humans.
“Language agents have demonstrated promising capabilities in automating web-based tasks, though their current reactive approaches still underperform largely compared to humans.
![Computer Science > Artificial Intelligence
arXiv:2411.06559 (cs)
[Submitted on 10 Nov 2024]
Is Your LLM Secretly a World Model of the Internet? Model-Based Planning for Web Agents
Yu Gu, Boyuan Zheng, Boyu Gou, Kai Zhang, Cheng Chang, Sanjari Srivastava, Yanan Xie, Peng Qi, Huan Sun, Yu Su
View PDF
HTML (experimental)
Language agents have demonstrated promising capabilities in automating web-based tasks, though their current reactive approaches still underperform largely compared to humans. While incorporating advanced planning algorithms, particularly tree search methods, could enhance these agents' performance, implementing tree search directly on live websites poses significant safety risks and practical constraints due to irreversible actions such as confirming a purchase. In this paper, we introduce a novel paradigm that augments language agents with model-based planning, pioneering the innovative use of large language models (LLMs) as world models in complex web environments. Our method, WebDreamer, builds on the key insight that LLMs inherently encode comprehensive knowledge about website structures and functionalities. Specifically, WebDreamer uses LLMs to simulate outcomes for each candidate action (e.g., "what would happen if I click this button?") using natural language descriptions, and then evaluates these imagined outcomes to determine the optimal action at each step. Empirical results on two representative web agent benchmarks with online interaction -- VisualWebArena and Mind2Web-live -- demonstrate that WebDreamer achieves substantial improvements over reactive baselines. By establishing the viability of LLMs as world models in web environments, this work lays the groundwork for a paradigm shift in automated web interaction. More broadly, our findings open exciting new avenues for future research into 1) optimizing LLMs specifically for world modeling in complex, dynamic environments, and 2) model-based speculative planning for language agents.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:552sltcbj2n32vqfg2zkjnyj/bafkreic6rrtzqqnmrg3sfqloujvm36zszohuis4bzrpy4zkug5mrzwx4dm@jpeg)
November 21, 2024 at 7:25 AM
This paper by Yu Gu et al looks *fascinating* arxiv.org/abs/2411.06559
“Language agents have demonstrated promising capabilities in automating web-based tasks, though their current reactive approaches still underperform largely compared to humans.
“Language agents have demonstrated promising capabilities in automating web-based tasks, though their current reactive approaches still underperform largely compared to humans.
