Sign in

Jaime Teevan

@teevan.bsky.social
1.3K followers 104 following 145 posts

Chief Scientist & Technical Fellow at Microsoft. Professor at UW. Mother to four wild boys. #AI #HCI #Productivity #FutureOfWork

PostsRepliesMedia
Jaime Teevan @teevan.bsky.social · 06/01/2025
How can AI move beyond simple responses and reliably address complex tasks? Meet Benjamin Van Durme (www.microsoft.com/en-us/resear...), who is pushing the boundaries of natural language tools so they can handle multi-step processes with contextual understanding.
170
Jaime Teevan @teevan.bsky.social · 26/11/2024
Can AI truly serve everyone, regardless of language or culture? Meet Sunayana Sitaram (www.linkedin.com/in/sunayana/), a Principal Researcher at Microsoft Research India, who is working hard to ensure that it will.
1131
Jaime Teevan @teevan.bsky.social · 23/11/2024
Curious about how AI and remote work are reshaping our workplaces? Meet Brent Hecht (www.linkedin.com/in/brenthecht/), a pioneer in human-centered AI research.
1151
Jaime Teevan @teevan.bsky.social · 12/02/2024
That’s a wrap! Hope the #NewFutureOfWork report helps you #LeadLikeAScientist. Of course that means experimenting. But remember, the scientific process is also about learning from others (e.g., via this report) and sharing & validating your findings. #Slide41 aka.ms/nfw2023pdf#p...
000
Jaime Teevan @teevan.bsky.social · 11/02/2024
Are LLMs the printing press of our time? They democratize knowledge, empower creativity, and enable collaboration. But they also require responsibility, accountability, and ethics. We can write the future of work with them. #NewFutureOfWork #Slide40 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 10/02/2024
An example of the importance of innovation v. automation via Erik Brynjolfsson: If the ancient Greeks had automated all labor sans innovation, nobody would have to work but we'd all still be using latrines. What will be the #GenAI equivalent to indoor plumbing? #Slide39 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 09/02/2024
I’m really curious about the new jobs AI will create. @donasarkar.bsky.social has some good ideas for the short-term around emerging AI-focused roles like AI validator, red-teamer, and adoption driver. But I bet my kids have jobs we haven’t even imagined yet. #Slide38 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 08/02/2024
Research consistently shows inexperienced workers receive more of a productivity boost from #genAI (#Slide8) - but it can only help them if they know about it. Currently people from counties with high rates of college education are more likely to search for #ChatGPT. #Slide37 aka.ms/nfw2023pdf#p...
110
Jaime Teevan @teevan.bsky.social · 07/02/2024
The dynamics of shared control between people and computers are complicated. Effective human-AI collab requires that AI systems be designed to support the human part when things fail. We need to engage with the entire sociotechnical system in which AI is deployed. #Slide36 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 06/02/2024
Monitoring my kids as toddlers was boring enough, monitoring AI is even worse! I developed #microproductivity to help myself multitask while watching my kids, but also explored its impact on self-driving car takeover: teevan.org/publications... #NewFutureOfWork #Slide35 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 05/02/2024
People form opinions about #GenAI – and anticipate the related workflow changes – before using it. Don't let your opinions stray from reality. Use AI to build your expectations from experience. Where to start? Copilot (copilot.microsoft.com)! #NewFutureOfWork #Slide34 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 04/02/2024
How do people react to new technologies? It depends on their expectations, hopes, fears. AI is no different. LLMs are part of a larger sociotechnical system that shapes and is shaped by people and contexts. How we design and use AI matters for our society and future. #Slide33 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 03/02/2024
"By bringing together people, knowledge, and working practice, ML knowledge bases can go beyond a consumption-oriented representation of past work, and instead become the medium through which work is conducted." www.microsoft.com/en-us/resear... by Lindley & Wilkins. #Slide32 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 02/02/2024
Microsoft came into existence as a document company. But LLMs can unlock the knowledge in docs through natural language. As we transition from a doc company to conversational one, the focus shifts to enabling great conversations (see: www.youtube.com/watch?v=f7Bd...). #Slide31 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 01/02/2024
If you don't reimagine how you work given the recent AI advances, you'll be left behind. But you shouldn't just think of AI as a new cog in your existing workflows. AI can track task interdependence, allocate roles, and redistribute work for better team outcomes. #Slide30 aka.ms/nfw2023pdf#p...
110
Jaime Teevan @teevan.bsky.social · 31/01/2024
What if LLMs could not only help us with our work, but also our work habits? By offering coaching and actionable feedback after a meeting, for example, LLMs can help us collaborate better. See MeetingCoach: dl.acm.org/doi/10.1145/... #NewFutureOfWork #Slide29 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 30/01/2024
Everyone knows Copilot summarizes meetings - so you can go to FEWER mtgs. But Copilot can also catalyze conversation - so you can go to BETTER mtgs. Look for suggested prompts like: "What Qs are unresolved?" or "List different perspectives by topic." #NewFutureOfWork #Slide28 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 29/01/2024
Imagine if an LLM turned a survey into an interview, asking you questions to draw out your best thoughts and then combining your insights with others. Feels like a new AI-mediated way to collaborate, which reminds me of the MicroWriter: teevan.org/publications.... #Slide27 aka.ms/nfw2023pdf#p...
110
Jaime Teevan @teevan.bsky.social · 28/01/2024
GPT4 aced the USMLE exam (required to practice medicine in the US), but can it actually help me stay healthy? I'd really like a standing metaprompt that subtly encourages me to eat well & exercise, even when not thinking about it. #NewFutureOfWork #Slide26 #GoalDirectedAI aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 27/01/2024
GitHub was our first product to integrate LLMs, and for good reason: LLMs are great at translating NL to code. Developers work 56% faster with AI, and those with less experience benefit most. Imagine the possibilities when coding's as easy as talking! #NewFutureOfWork #Slide24 aka.ms/nfw2023pdf#p...
110
Jaime Teevan @teevan.bsky.social · 26/01/2024
You can’t perceive a 100ms delay, but when your search results are just 100ms slower you think they’re worse. LLMs prove the benefit of little extra compute time. Can #SlowSearch (teevan.org/publications...) teach us how to make the latency worth it? #NewFutureofWork #Slide23 aka.ms/nfw2023pdf#p...
110
Jaime Teevan @teevan.bsky.social · 25/01/2024
How did hashtags and at-mentions become a thing? Twitter learned from user behavior. LLMs are no different. Millions of chats are creating new patterns and possibilities every day. What new features will emerge? #NewFutureofWork #Slide22 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 24/01/2024
I used to think AI would do my grunt work, learning the things I do over and over again. But LLMs have surprised me in their value for creative work. We even used one in the early Copilot days to generate ideas for unexpected ways to integrate LLMs into our products. #Slide21 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 23/01/2024
Don't let AI go to your head! GenAI imposes new metacognitive demands, requiring we plan, monitor & evaluate our cognitive strategies. One way to improve AI systems is to explicitly integrate metacognitive support (e.g., via self-reflective prompts). #NewFutureofWork #Slide19 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 22/01/2024
LLMs will change the skills students need to learn, and the way they learn them. I look forward to exploring how AI can support (v. substitute for) learning in my keynote at #BETT2024 this Wed (uk.bettshow.com/keynote-spea...). #NewFutureofWork #Slide25 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 21/01/2024
Parlez-vous IA? LLMs do surprisingly well with different languages. So well, in fact, that Copilot is the first time Microsoft has released a major AI feature that works in all of our most popular languages right out of the gate. #NewFutureofWork #Slide20 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 20/01/2024
You need to audit AI output differently from human output because it makes different kinds of mistakes. E.g., AI is unlikely to be adversarial but more likely to hallucinate. Co-audit tools (arxiv.org/pdf/2310.012...) help with evaluating AI output. #NewFutureofWork #Slide18 aka.ms/nfw2023pdf#p...
101
Jaime Teevan @teevan.bsky.social · 19/01/2024
You can trust AI more when you know when it's wrong. Spatharioti et al. (arxiv.org/pdf/2307.037...) showed that highlighting uncertain AI answers improved decision accuracy. Uncertainty can be a chance to learn from AI, not just trust it. #NewFutureofWork #Slide17 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 18/01/2024
How to avoid falling for AI's mistakes: Use a “cognitive forcing function” to actively evaluate the AI and your own judgment. For example, explain to yourself why you agree or disagree with the response you get. More tips: www.microsoft.com/en-us/resear.... #NFW #Slide16 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 17/01/2024
Uncertainty’s not a bug, it’s a feature, enabling human-AI collaboration that’s fluid and adaptive. Mixed-initiative computing (erichorvitz.com/ftp/mixedin....) leverages uncertainty to design AI that aligns with human values and preferences. #Slide15 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 16/01/2024
As you try out the new Copilot Pro, prompt advice abounds (e.g., hbr.org/2023/12/to-w...). But even better than having to figure it out yourself, prompt support is starting to be built directly into #genAI tools. E.g., Copilot Lab: copilot.cloud.microsoft/prompts. #Slide14 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 15/01/2024
Prompt engineering is a hot topic in ML research. Many papers find a good prompt on a weak model can outperform a bad one on better model (e.g., arxiv.org/abs/2311.16452). To make these gains durable, researchers now use LLMs for prompt writing, too. #NewFutureofWork #Slide13 aka.ms/nfw2023pdf#p...
110
Jaime Teevan @teevan.bsky.social · 14/01/2024
DYK adding "Let's work together to solve this problem!" to your prompt can improve the outcome (arxiv.org/abs/2309.03409)? But don't get too attached to your prompts – LLMs are evolving fast so your favorites might not work as well on a new model. #NewFutureofWork #Slide12 aka.ms/nfw2023pdf#p...
110
Jaime Teevan @teevan.bsky.social · 13/01/2024
With 4 kids in (or soon to start) college, I sometimes wonder what skills they'll need as work changes. I hope I'm teaching them critical thinking and emotional intelligence - but, of course, that's hard to tell with teenagers! :P #NewFutureofWork #Slide11 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 12/01/2024
#AIOrchestration is about managing LLMs to optimize their performance. Microproductivity is about helping people manage complex tasks for similar ends. Both coordinate work by using AI to plan, track, and synthesize it. #Microproductivity == #HumanOrchestration? #Slide10 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 11/01/2024
My favorite way to use an LLM is for critical thinking. When summarizing an article, I prompt, “What questions should I ask?” When writing a hard email, I ask it to predict how my message will land. I even used an LLM to get feedback on this post. #NewFutureofWork #Slide9 aka.ms/nfw2023pdf#p...
110
Jaime Teevan @teevan.bsky.social · 10/01/2024
In a significant departure from previous info technologies, inexperienced workers receive more of the productivity boost from generative AI. This shows up across multiple studies by researchers like Brynjolfsson, Dell'Acqua, Mollick, Noy, Peng, etc. #NewFutureofWork #Slide8 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 09/01/2024
When you first try using AI for work, start with the tasks where the research literature shows it'll help the most, like search, summarization, first drafts. According to aka.ms/productivity..., you'll see real productivity gains. #NewFutureofWork #Slide7 aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 08/01/2024
Most research right now studies how much faster AI makes you (answer: a lot, with similar quality output). But I'm more curious how people will use the extra time. To produce unimaginably high quality? Do entirely new things? #NewFutureofWork #Slide6 #GoDawgs aka.ms/nfw2023pdf#p...
100
Jaime Teevan @teevan.bsky.social · 07/01/2024
There is a LOT being published right now about how LLMs are changing work. The #NewFutureofWork report focuses specifically on areas that we think deserve additional attention or where research being done at Microsoft offers a unique perspective. #Slide5 aka.ms/nfw2023
110
Jaime Teevan @teevan.bsky.social · 06/01/2024
I've never seen so many different orgs at Microsoft (MSR, product, marketing, strategy, HR, facilities) collaborate on research as with the #NewFutureofWork initiative (aka.ms/nfw). And the research has had broad impact: on product, internal policy, and externally. #Slide4 aka.ms/nfw2023
110
Jaime Teevan @teevan.bsky.social · 05/01/2024
Since 2020, work has changed radically! The 2021 #NewFutureofWork report (aka.ms/nfw2021) focused on remote work, 2022 on hybrid (aka.ms/nfw2022), and now 2023 on AI (aka.ms/nfw2023). These shifts are a progression: AI for work requires its digitization. #Slide3
100
Jaime Teevan @teevan.bsky.social · 04/01/2024
The #NewFutureofWork report from Microsoft Research covers research by hundreds of scientists around the world, compiled by 59 authors and edited by Jenna, Sonia, @nancybaym.bsky.social, Mary, Shamsi, Kate, Sean, Abigail, Mickey, Brent, and me. #Slide2 aka.ms/nfw2023
121
Jaime Teevan @teevan.bsky.social · 03/01/2024
The #NewFutureofWork report synthesizes new – and newly relevant – academic research related to AI's impact on work. But it's too much to read in just one day! So I'm going to post a slide-a-day, highlighting tidbits and providing links to learn more. #Slide1 aka.ms/nfw2023
130
Jaime Teevan @teevan.bsky.social · 21/08/2023
"I'm not interested in when [AI] works," says chief scientist Teevan. "I'm like, 'Show me when it doesn't work.' And Satya is very much in that .. mindset: 'Let's find the cases where it breaks and how we're learning from that.'" fastcompany.com/90931084/satya-nade…
Headshot of Jaime Teevan with a blue background.
191