Juan Mateos-Garcia
@jmateosgarcia.bsky.social
Impact data analyst at Google DeepMind. Interested in AI, economics, complexity, metascience, research methods, science-fiction.
"Artificial Intelligence and the Labor Market"
Nice analysis of LinkedIn data to estimate the impact of AI on jobs between 2010 and 2023.
It illustrates the complexity of AI economic impacts: what exposure taketh away, complementarities and productivity giveth.
www.menakahampole.com
Nice analysis of LinkedIn data to estimate the impact of AI on jobs between 2010 and 2023.
It illustrates the complexity of AI economic impacts: what exposure taketh away, complementarities and productivity giveth.
www.menakahampole.com
February 26, 2025 at 12:09 PM
"Artificial Intelligence and the Labor Market"
Nice analysis of LinkedIn data to estimate the impact of AI on jobs between 2010 and 2023.
It illustrates the complexity of AI economic impacts: what exposure taketh away, complementarities and productivity giveth.
www.menakahampole.com
Nice analysis of LinkedIn data to estimate the impact of AI on jobs between 2010 and 2023.
It illustrates the complexity of AI economic impacts: what exposure taketh away, complementarities and productivity giveth.
www.menakahampole.com
Metascience and AI postdoctoral fellowship
Cool @sloanfoundation.bsky.social opportunity for anyone interested in studying AI x science adoption and impact. It is also very aligned with the evidence and experimentation recommendation in our recent essay!
sloan.org/programs/dig...
Cool @sloanfoundation.bsky.social opportunity for anyone interested in studying AI x science adoption and impact. It is also very aligned with the evidence and experimentation recommendation in our recent essay!
sloan.org/programs/dig...
February 14, 2025 at 1:46 PM
Metascience and AI postdoctoral fellowship
Cool @sloanfoundation.bsky.social opportunity for anyone interested in studying AI x science adoption and impact. It is also very aligned with the evidence and experimentation recommendation in our recent essay!
sloan.org/programs/dig...
Cool @sloanfoundation.bsky.social opportunity for anyone interested in studying AI x science adoption and impact. It is also very aligned with the evidence and experimentation recommendation in our recent essay!
sloan.org/programs/dig...
Could much of the recent drama about DeepSeek be rooted on flawed causal thinking i.e. neglecting the counterfactual? Dario Amodei makes a compelling case here.
darioamodei.com/on-deepseek-...
darioamodei.com/on-deepseek-...
January 29, 2025 at 6:53 PM
Could much of the recent drama about DeepSeek be rooted on flawed causal thinking i.e. neglecting the counterfactual? Dario Amodei makes a compelling case here.
darioamodei.com/on-deepseek-...
darioamodei.com/on-deepseek-...
They also forecast skill mismatches based on the network structure of job impacts e.g. whether a job’s neighbours see higher demand (bad for employers) or lower demand (bad for workers). This could inform timely policies to accelerate deployment and reduce disruption.
January 29, 2025 at 9:39 AM
They also forecast skill mismatches based on the network structure of job impacts e.g. whether a job’s neighbours see higher demand (bad for employers) or lower demand (bad for workers). This could inform timely policies to accelerate deployment and reduce disruption.
They find that the net impact of rapid decarbonisation on jobs is positive but depends on the stage of the deployment cycle. Many jobs see growth during the “scale up phase” (build up of low-carbon infrasructure) but decline afterwards.
January 29, 2025 at 9:39 AM
They find that the net impact of rapid decarbonisation on jobs is positive but depends on the stage of the deployment cycle. Many jobs see growth during the “scale up phase” (build up of low-carbon infrasructure) but decline afterwards.
They forecast the impact of energy tech adoption scenarios on output and labour demand in exposed sectors, and the implications for employers and workers based on impacts in related occupations.
January 29, 2025 at 9:39 AM
They forecast the impact of energy tech adoption scenarios on output and labour demand in exposed sectors, and the implications for employers and workers based on impacts in related occupations.
"Employment dynamics in a rapid decarbonization of the US power sector"
Cool complexity economics analysis of a scenario that sadly looks a bit sci-fi in the current political circumstances. It would be interesting to study labour market impacts of AI with this approach.
Cool complexity economics analysis of a scenario that sadly looks a bit sci-fi in the current political circumstances. It would be interesting to study labour market impacts of AI with this approach.
January 29, 2025 at 9:39 AM
"Employment dynamics in a rapid decarbonization of the US power sector"
Cool complexity economics analysis of a scenario that sadly looks a bit sci-fi in the current political circumstances. It would be interesting to study labour market impacts of AI with this approach.
Cool complexity economics analysis of a scenario that sadly looks a bit sci-fi in the current political circumstances. It would be interesting to study labour market impacts of AI with this approach.
They analyse the distribution of suspense and surprise in technology codes within higher level technologies and characterise a technology lifecycle of surprise - suspense - popularity. This is what it looks like for AI. It would be interesting to drill down further!
January 27, 2025 at 7:49 AM
They analyse the distribution of suspense and surprise in technology codes within higher level technologies and characterise a technology lifecycle of surprise - suspense - popularity. This is what it looks like for AI. It would be interesting to drill down further!
They show that suspenseful innovations tend to receive more citations and be commercially valuable. They speculate that this could be because society is more prepared to adopt suspenseful innovations than surprising ones (this underscores the value of tech forecasting / DARPAs)
January 27, 2025 at 7:49 AM
They show that suspenseful innovations tend to receive more citations and be commercially valuable. They speculate that this could be because society is more prepared to adopt suspenseful innovations than surprising ones (this underscores the value of tech forecasting / DARPAs)
Using patents, they create a model to predict if two technologies will be connected. If their model predicts the connection before it happens it is suspenseful. If not, it is surprising.
January 27, 2025 at 7:49 AM
Using patents, they create a model to predict if two technologies will be connected. If their model predicts the connection before it happens it is suspenseful. If not, it is surprising.
"Suspense and surprise in technological innovation"
Neat analysis of the link between expectations and outcomes in technological evolution.
Neat analysis of the link between expectations and outcomes in technological evolution.
January 27, 2025 at 7:49 AM
"Suspense and surprise in technological innovation"
Neat analysis of the link between expectations and outcomes in technological evolution.
Neat analysis of the link between expectations and outcomes in technological evolution.
Charles Dickens on Victorian UBI (from Bleak House
January 25, 2025 at 9:39 AM
Charles Dickens on Victorian UBI (from Bleak House
This results in a trade-off: These 12 elements benefit from co-location but requier independence.
The essay argues that Bell labs balanced them by retaining a focus on advancing knowledge while staying connected to AT&Ts technical / commercial problems.
The essay argues that Bell labs balanced them by retaining a focus on advancing knowledge while staying connected to AT&Ts technical / commercial problems.
January 24, 2025 at 7:44 AM
This results in a trade-off: These 12 elements benefit from co-location but requier independence.
The essay argues that Bell labs balanced them by retaining a focus on advancing knowledge while staying connected to AT&Ts technical / commercial problems.
The essay argues that Bell labs balanced them by retaining a focus on advancing knowledge while staying connected to AT&Ts technical / commercial problems.
None of these components is superior / foundational. Science helps advance technology and technology helps advance science. Both can be incremental or exploratory.
Incentives / institutions create a bias towards tackling existing questions and problems (vs. exploring new questions / problems.)
Incentives / institutions create a bias towards tackling existing questions and problems (vs. exploring new questions / problems.)
January 24, 2025 at 7:44 AM
None of these components is superior / foundational. Science helps advance technology and technology helps advance science. Both can be incremental or exploratory.
Incentives / institutions create a bias towards tackling existing questions and problems (vs. exploring new questions / problems.)
Incentives / institutions create a bias towards tackling existing questions and problems (vs. exploring new questions / problems.)
Their model has 12 components (a "dodecant"):
2 areas: science and technology
3 mechanisms for progress in each area (facts/explanations/generalisations and functions/forms/exaptions)
2 styles of progress: generate surprises / consolidate knowledge
2 areas: science and technology
3 mechanisms for progress in each area (facts/explanations/generalisations and functions/forms/exaptions)
2 styles of progress: generate surprises / consolidate knowledge
January 24, 2025 at 7:44 AM
Their model has 12 components (a "dodecant"):
2 areas: science and technology
3 mechanisms for progress in each area (facts/explanations/generalisations and functions/forms/exaptions)
2 styles of progress: generate surprises / consolidate knowledge
2 areas: science and technology
3 mechanisms for progress in each area (facts/explanations/generalisations and functions/forms/exaptions)
2 styles of progress: generate surprises / consolidate knowledge
"How technoscientific knowledge advances"
The genesis of TS revolutions was one of the best innovation books in recent years. In this new essay, the authors summarise and expand its ideas.
The genesis of TS revolutions was one of the best innovation books in recent years. In this new essay, the authors summarise and expand its ideas.
January 24, 2025 at 7:44 AM
"How technoscientific knowledge advances"
The genesis of TS revolutions was one of the best innovation books in recent years. In this new essay, the authors summarise and expand its ideas.
The genesis of TS revolutions was one of the best innovation books in recent years. In this new essay, the authors summarise and expand its ideas.
Christmas reading/re-reading ambitions.
December 20, 2024 at 5:11 PM
Christmas reading/re-reading ambitions.
AI is going to transform how I engage with theoretical economics papers.
December 19, 2024 at 8:16 AM
AI is going to transform how I engage with theoretical economics papers.
The report offers a glimpse into the kinds of economics research this data could enable e.g. track AI adoption in different tasks, occupations, industries and locations, nowcast real-world outcomes through AI usage in real-time as has been done with Google Trends, and more...
December 17, 2024 at 7:49 AM
The report offers a glimpse into the kinds of economics research this data could enable e.g. track AI adoption in different tasks, occupations, industries and locations, nowcast real-world outcomes through AI usage in real-time as has been done with Google Trends, and more...
They have built an interactive tool where the trust and safety team can explore the clusters at different levels of resolution to identify suspicious behaviours (worth noting that they are careful to preserve user privacy at all steps of the analysis).
December 17, 2024 at 7:49 AM
They have built an interactive tool where the trust and safety team can explore the clusters at different levels of resolution to identify suspicious behaviours (worth noting that they are careful to preserve user privacy at all steps of the analysis).
An interesting question they don't discuss is how they maintain / update their hierarchical clustering longitudinally to track cluster evolution and emergence. This is how we thought this could be done in the context of industrial taxonomies:
December 17, 2024 at 7:49 AM
An interesting question they don't discuss is how they maintain / update their hierarchical clustering longitudinally to track cluster evolution and emergence. This is how we thought this could be done in the context of industrial taxonomies:
Their clustering pipeline is neat: they use embeddings and K-means to cluster conversations at the lowest level of resolution and Claude aggregates and labels clusters hierarchically. Apparently this outperforms traditional hierarchical clustering algorithms!
December 17, 2024 at 7:49 AM
Their clustering pipeline is neat: they use embeddings and K-means to cluster conversations at the lowest level of resolution and Claude aggregates and labels clusters hierarchically. Apparently this outperforms traditional hierarchical clustering algorithms!
Clio: Privacy-preserving insights into Real-World AI use.
Anthropic have developed a pipeline to cluster and visualise Claude user conversations. This is informing trust and safety efforts and creating evidence about AI real-world adoption. Very cool!
assets.anthropic.com/m/7e1ab885d1...
Anthropic have developed a pipeline to cluster and visualise Claude user conversations. This is informing trust and safety efforts and creating evidence about AI real-world adoption. Very cool!
assets.anthropic.com/m/7e1ab885d1...
December 17, 2024 at 7:49 AM
Clio: Privacy-preserving insights into Real-World AI use.
Anthropic have developed a pipeline to cluster and visualise Claude user conversations. This is informing trust and safety efforts and creating evidence about AI real-world adoption. Very cool!
assets.anthropic.com/m/7e1ab885d1...
Anthropic have developed a pipeline to cluster and visualise Claude user conversations. This is informing trust and safety efforts and creating evidence about AI real-world adoption. Very cool!
assets.anthropic.com/m/7e1ab885d1...
The GenCast paper is very cool. I specially loved the analysis of the relative economic value of different weather prediction models. This shows that GenCast could improve efficiency by helping decision-makers prepare for extreme weather events when they need to.
www.nature.com/articles/s41...
www.nature.com/articles/s41...
December 10, 2024 at 10:55 AM
The GenCast paper is very cool. I specially loved the analysis of the relative economic value of different weather prediction models. This shows that GenCast could improve efficiency by helping decision-makers prepare for extreme weather events when they need to.
www.nature.com/articles/s41...
www.nature.com/articles/s41...