Sabitlenmiş Tweet
Greg Leppert
1.8K posts

Greg Leppert
@leppert
Working on AI and access to knowledge. Executive Director @instdin, Chief Technologist @BKCHarvard
Most tweets deleted after 1 week. Katılım Şubat 2010
648 Takip Edilen2.4K Takipçiler

@DavidDuvenaud @AlecRad @status_effects @instdin But if your goal is testing LMs for their ability to predict the future, suddenly there’s reason to care about the qualities of data that knowledge stewards have long valued. Suddenly everyone is working together on the same problems.
English

@DavidDuvenaud @AlecRad @status_effects @instdin What’s particularly smart is this team’s aligning of incentives. It’s been difficult to get the AI community to care about data quality beyond hill climbing toward benchmarks perceived as capturing the sota of model behavior.
English

Announcing Talkie: a new, open-weight historical LLM! We trained and finetuned a 13B model on a newly-curated dataset of only pre-1930 data. Try it below!
with @AlecRad and @status_effects 🧵
English
Greg Leppert retweetledi

@DavidDuvenaud @AlecRad @status_effects @instdin It shouldn’t be missed the profound dedication to data cleanliness and accuracy here. Everyone in the AI community working on and with historical data should take note. This is how we germinate force multipliers between the work of AI builders, historians, and digital humanists.
English

If you’re interested in working with us at @instdin to produce state of the art datasets in collaboration with knowledge institutions from across the globe, reach out. We’re hiring deep technologists and community builders. institutional.org
English

Amazing work from an amazing team using @instdin’s Institutional Books data release. Their dedication to detail and accuracy is sorely missing from the vast majority of historical-data work from the AI community. Yet there’s so much work to be done and benefit to getting it right
David Duvenaud@DavidDuvenaud
Announcing Talkie: a new, open-weight historical LLM! We trained and finetuned a 13B model on a newly-curated dataset of only pre-1930 data. Try it below! with @AlecRad and @status_effects 🧵
English

@markankcorn @BWarburg @Winterrose @HarvardLIL In our experience, most people didn’t want or need an API—they either wanted wholesale data dumps or to browse specific cases via a GUI. Most API use fell into the former.
English

The first child teaches you about yourself. The second child teaches you about the first.
Lionel Page@page_eco
All parents think their parenting shapes their child until they have a second child. Then they realise it was the child’s personality all along. (Marvin Zuckerman)
English
Greg Leppert retweetledi






