Michelle Lam

@michelle123lam

Incoming assistant professor @UTAustin CS | PhD @Stanford CS | hci, human-ai interaction (+ dance, design, doodling!)

2 363Followers

731Following

187Posts total

114.4KViews on collected posts

Últimas publicaciones

Michelle Lam@michelle123lam15 Apr

@michelle123lam Congratulations Michelle! I’ve been following this work since it was first posted on arxiv. I really like it and share it with many friends! Regretfully I didn’t make it to CHI. But I hope that I can have some opportunities to catch up with you in the future! 581 views · 3 likes · 0 reposts · 1 replies Open on X →

Michelle Lam@michelle123lam15 Apr

In hour-long in-lab sessions where users applied the system to their own writing tasks (N=17), Poppins produced specialized outputs that are unique to each participant and are rated as significantly higher quality than that of a standard LLM chat tool. 1.5K views · 12 likes · 0 reposts · 1 replies Open on X →

Michelle Lam@michelle123lam15 Apr

Thanks so much to co-authors @oshaikh13, Hallie Xu, @azguo2, @Diyi_Yang, @jeffrey_heer, @landay, & @msbernst! We’re grateful that this work received a CHI best paper honorable mention! paper: https://t.co/jRh5aDyTuM waitlist: https://t.co/4PsdjtKpgp talk: Th 4/16 12pm @ P1-111 3K views · 40 likes · 3 reposts · 0 replies Open on X →

Michelle Lam@michelle123lam15 Apr

JIT objectives power Poppins, a browser extension for on-demand UI generation. Based on a user’s screen, the system can induce a JIT objective and then produce tool ideas, tool specs, and tool code to render specialized software on the fly: all without a single prompt required! h 884 views · 13 likes · 0 reposts · 1 replies Open on X →

Michelle Lam@michelle123lam15 Apr

This is possible because JIT objectives are highly accurate: 77% were rated Accurate/Very Accurate, and JIT objectives were selected over a manually-written objective in 98% of cases. In addition, JIT objective weights were aligned with participants’ relative preferences. https:/ 972 views · 10 likes · 0 reposts · 1 replies Open on X →

Michelle Lam@michelle123lam15 Apr

Once you have JIT objectives, you can embed them into various LLM architectures via existing generators and evaluators. Evaluations on N=205 participant-provided inputs show that JIT objectives produce user-preferred outputs, whether generating experts, tools, or feedback. https: 10.1K views · 16 likes · 0 reposts · 1 replies Open on X →

Michelle Lam@michelle123lam15 Apr

At interaction time, we have a surprising number of cues such that we can induce detailed objectives without any user effort. By observing user interaction traces, *Just-In-Time Objectives* model the user’s in-the-moment goals to specialize LLM systems on the fly. https://t.co/1d 1.8K views · 17 likes · 1 reposts · 1 replies Open on X →

Michelle Lam@michelle123lam15 Apr

Most of what I actually need help with, I never think to tell a model. But why is it on me to remember? Our new paper asks: what if AI could proactively specialize to individuals and the tasks they’re carrying out at this very moment? 🧵 https://t.co/Mt6jf0V0Qq 50.4K views · 279 likes · 43 reposts · 13 replies Open on X →

Michelle Lam@michelle123lam19 Apr 2024

@michelle123lam this is really cool and timely, Michelle! 255 views · 1 likes · 0 reposts · 1 replies Open on X →

Michelle Lam@michelle123lam18 Apr 2024

In evals, LLooM exceeds baselines to recover 70-90% of human-annotated, generic topics, and LLooM is significantly better at surfacing specific, nuanced concepts. From content moderation to AI ethics statements, LLooM concepts help us make sense of data: https://t.co/P1phTRBCpW 675 views · 6 likes · 0 reposts · 1 replies Open on X →

Michelle Lam@michelle123lam18 Apr 2024

Going further, expert data analysts used LLooM to uncover novel insights even on familiar datasets. Analysts were particularly excited that LLooM facilitated theory-driven analysis: they could read out patterns and write hypotheses through the language of LLooM concepts. https:// 1.2K views · 7 likes · 0 reposts · 2 replies Open on X →

Michelle Lam@michelle123lam18 Apr 2024

This work would not be possible without my amazing collaborators and advisors—Janice Teoh, @landay, @jeffrey_heer, and @msbernst! Start using LLooM to explore text data via concepts (https://t.co/DdGlDitDKy), or see our #CHI2024 paper to learn more (https://t.co/pRV0S7ECuc)! 1.3K views · 19 likes · 1 reposts · 3 replies Open on X →

Michelle Lam@michelle123lam18 Apr 2024

We instantiate our algorithm in the LLooM Workbench, an open-source text analysis tool for computational notebooks. Users can explore text data in terms of high-level concepts—from a dataset overview to concept details to document-level scores, highlights, and rationale. https:// 711 views · 6 likes · 0 reposts · 1 replies Open on X →

Michelle Lam@michelle123lam18 Apr 2024

Our algorithm draws on the ability of large language models (LLMs) to generalize from examples. LLooM samples extracted text and iteratively synthesizes proposed concepts of increasing generality. Once concepts are induced, LLooM can classify the entire dataset. https://t.co/JYGt 778 views · 8 likes · 0 reposts · 1 replies Open on X →

Michelle Lam@michelle123lam18 Apr 2024

LLooM fills this gap with concept induction, extracting high-level concepts defined by natural language descriptions & explicit inclusion criteria (e.g., “Dismissal of women’s concerns: Does the text dismiss or invalidate women’s concerns or experiences?”) https://t.co/DdGlDitDKy 923 views · 7 likes · 0 reposts · 1 replies Open on X →

Michelle Lam@michelle123lam18 Apr 2024

“Can we get a new text analysis tool?” “No—we have Topic Model at home” Topic Model at home: outputs vague keywords; needs constant parameter fiddling🫠 Is there a better way? We introduce LLooM, a concept induction tool to explore text data in terms of interpretable concepts🧵 h 38.2K views · 203 likes · 39 reposts · 4 replies Open on X →

Michelle Lam@michelle123lam18 Apr 2024

Analysts have questions like “How are women in power described?” Vague topics like “women, power, female” aren’t what they’re after—they want to understand data with nuanced concepts like “Criticism of traditional gender roles,” which are central to theory-driven data analysis. 1.1K views · 4 likes · 0 reposts · 2 replies Open on X →

Growth & engagement

How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.

Views per post

77818 Apr

711

1.3K

1.2K

675

25519 Apr

50.4K15 Apr

1.8K

10.1K

972

884

1.5K

581

Last 14 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

1.16%18 Apr

0.98%

1.76%

0.72%

1.04%

0.78%19 Apr

0.68%15 Apr

1.07%

0.18%

1.13%

1.58%

1.43%

0.87%

0.69%

Reactions — likes, reposts, replies and quotes — divided by views.

What the audience does

Likes52.8%651 in total

Reposts7.1%87 in total

Replies2.8%35 in total

Quotes1.3%16 in total

Bookmarks36.1%445 in total

Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.

The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.

Últimas publicaciones

Growth & engagement

Views per post

Engagement rate per post

What the audience does

Cuentas similares