Amy Heger (@AmyHeger) header
Amy Heger (@AmyHeger) profile picture

Amy Heger

@AmyHeger

UX Researcher @jeffrymconsulting | Human-centered and Responsible AI | Social Psychologist | Opinions are my own

 St. Louis, MO
linkedin.com/in/amy-heger
Joined May 2010
  • Tweets893
  • Following1.3K
  • Followers338
114Media
Photo by Amy Heger (@AmyHeger)Photo by Amy Heger (@AmyHeger)Photo by Amy Heger (@AmyHeger)Photo by Amy Heger (@AmyHeger)Photo by Amy Heger (@AmyHeger)Photo by Amy Heger (@AmyHeger)Photo by Amy Heger (@AmyHeger)Photo by Amy Heger (@AmyHeger)Photo by Amy Heger (@AmyHeger)
You’d think these requirements would’ve existed earlier based on the number of hazing-related deaths over the years, but hooray for Biden signing an anti-hazing bill that mandates institutions to share hazing statistics. wapo.st/41VLjF0
1
1
41
New from my team: Synthesis of research on appropriate reliance on generative AI, linked in the MSR blog below. @samirpassi and I will post interesting insights in the coming days.
Microsoft Research (@MSFTResearch) profile pictureMicrosoft Research
@MSFTResearch17 Apr 2024
In this issue: New research on appropriate reliance on generative AI; Power management opportunities for LLMs in the cloud; LLMLingua-2 improves task-agnostic prompt compression; Enhancing COMET to embrace under-resourced African languages. msft.it/6010Y6hCg
Decorative graphic with blue, purple, and pink gradients and abstract shapes. White text overlay reads: "Research Focus: April 15, 2024"

ALT Decorative graphic with blue, purple, and pink gradients and abstract shapes. White text overlay reads: "Research Focus: April 15, 2024"

3
1
12
4,299
Replying to @mustafasuleyman @Microsoft
Welcome to Microsoft! Let's talk about building appropriate reliance on our consumer products - so people don't over-trust AI when it's wrong, and don't under-trust either. I lead a cross-company team on this.
1
7
460
Lex is right, Sam was wrong. Our research shows that people don’t really check the work of AIs, once those AIs cross a certain threshold of quality. We found that consultants, for example, “fell asleep at the wheel” and did not look for AI errors. Training didn’t help, either.
Image from Ethan Mollick (@emollick)
64
189
50
1,168
163,375
The RAI Maturity Model is the most meaningful research project I've worked on. Prioritizing collaboration and communication across disciplines is essential for any org that wants to build AI responsibly. Read more at @mihaela_v's LinkedIn #RAI #maturity
Mickey Vorvoreanu (Dr. V) (@mihaela_v) profile pictureMickey Vorvoreanu (Dr. V)@mihaela_v15 Mar 2024
Y'all, I've been doing a series of posts about the RAI Maturity Model on LinkedIn linkedin.com/in/mihaelavorvo… Today: building a *common language* across disciplines so you can work together.
2
9
1,056
PhD student internship opportunity - work with our team and MS Office to research GenAI/Copilot usage!
Mickey Vorvoreanu (Dr. V) (@mihaela_v) profile pictureMickey Vorvoreanu (Dr. V)@mihaela_v30 Jan 2024
Internship for PhD students: jobs.careers.microsoft.com/g… - Work with my team & with MS Office. I am looking for a social scientist with experience in organizational communication, new employee experience, productivity, mixed-methods field studies.
1
195
Aether, founded by @erichorvitz and @BradSmi, was the first group to start thinking about Responsible AI at @Microsoft. Our deep thinking continues to influence the company's approach to RAI. See what's top of mind for Aether in Review of 2023 research: aka.ms/AAoiqqy
3
7
3
25
9,290
Technology Review's perspective on Google's Gemini model, with comments from @chirag_shah, @emilymbender, @percyliang, and me. technologyreview.com/2023/12…
Google DeepMind’s new Gemini model looks amazing—but could signal peak AI hype

Google DeepMind’s new Gemini model looks amazing—but could signal peak AI hype

It outmatches GPT-4 in almost all ways—but only by a little. Was the buzz worth it?

technologyreview.com
11
47
2
231
38,850
Proud to share Microsoft's AI & Productivity research initiative's release of its first public report. The research measures the impact AI-powered tools like Copilot have on productivity and we saw drastic gains in time saved on common info worker tasks tinyurl.com/7dfkvtsj
91
Interesting work on how the power of prompting can lead a generalist LLM to be a specialist, outperforming a fine-tuned model for medical applications. New great research coming from people on my team at Microsoft.
Eric Horvitz (@erichorvitz) profile pictureEric Horvitz
@erichorvitz29 Nov 2023
1/8 We’ve published a study of the power of prompting to unleash expertise from GPT-4 on medical benchmarks without additional fine-tuning or expert-curated prompts: arxiv.org/pdf/2311.16452.pdf Summary of results:
Image from Eric Horvitz (@erichorvitz)
1
197
if you’re not in tech then all you need to know about the past weekend is that your tech friends went through like three seasons of succession
54
1,150
144
11,675
904,017
We’re thrilled to present our new Focus on artificial intelligence - Navigating the AI Frontier. (1/20) buff.ly/3upNIch
2
61
5
115
75,081
Join me Wednesday as I talk with @HeyMarvinApp about how my colleagues at Microsoft & I created the Responsible AI Maturity Model. Hint: it involved analyzing a huge amount of complex qualitative data and I'll share many of our learnings. Register at: tinyurl.com/5ff2d5nm
2
3
1
9
1,507
I also want to shout out to @mihaela_v Shipi Dhanorkar @samirpassi @RuotongWang1 Zoe Kahn for all your work helping develop the Responsible AI Maturity Model aka.ms/raimm and to the 90+ RAI consultants & AI practitioners we interviewed that made its creation possible!
1
4
128
Adding to this how-to steps for RAI red-teaming from our team at Microsoft learn.microsoft.com/en-us/az…
Planning red teaming for large language models (LLMs) and their applications - Microsoft Foundry

Planning red teaming for large language models (LLMs) and their applications - Microsoft Foundry

Learn about how red teaming and adversarial testing are an essential practice in the responsible development of systems and features using large language models (LLMs)

learn.microsoft.com
MMitchell (@mmitchell_ai) profile pictureMMitchell
@mmitchell_ai21 Sep 2023
Interestingly, now that #OpenAI has announced their red-teaming effort, I'm being asked even more: "WHAT IS REDTEAMING?" This blog post from @nazneenrajani @natolambert @_lewtun @huggingface should be helpful! huggingface.co/blog/red-team…
Red-Teaming Large Language Models

Red-Teaming Large Language Models

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co
3
3
877
This is still the best AI ever generated Salmon jumping in a river
Image from Chris Albon (@chrisalbon)
Image from Chris Albon (@chrisalbon)
Image from Chris Albon (@chrisalbon)
214
4,539
283
54,995
3,156,761
Zoom updated its Terms of Service to permit training AI on user content without opt-out, a thread 🧵
Image from Bobi Rakova (@bobirakova)
16
280
49
606
128,305
New paper: Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback We survey over 250 papers to review challenges with RLHF with a focus on large language models. Highlights in thread 🧵
Image from Cas (Stephen Casper) (@StephenLCasper)
13
156
29
690
200,380
Today is the day that the world's first algorithmic bias auditing law goes into effect in NYC! Has anyone seen one in the wild yet? The law does not exactly make it easy to find them. nbcnews.com/tech/tech-news/n…
In NYC, companies will have to prove their AI hiring software isn't sexist or racist

In NYC, companies will have to prove their AI hiring software isn't sexist or racist

AI-infused hiring programs have drawn scrutiny, most notably over whether they end up exhibiting biases based on the data they’re trained on.

nbcnews.com
5
31
2
111
12,735
Woohooo! Our RAI Maturity Model is featured in this update. Find it at aka.ms/raimm @AmyHeger @samirpassi
Microsoft Research (@MSFTResearch) profile pictureMicrosoft Research
@MSFTResearch26 Jun 2023
In this issue: Our new Responsible AI Maturity Model; FoundWright helps “re-find” web pages; Trace-guided Inductive Synthesis of Recursive Functional Programs; a wait-free algorithm for weak reference counting; and new research on concurrency testing. msft.it/6019gscCv
Research Focus: Responsible AI Maturity Model, “re-finding” web pages

Research Focus: Responsible AI Maturity Model, “re-finding” web pages

In this issue: Our new Responsible AI Maturity Model; FoundWright helps “re-find” web pages; Trace-guided Inductive Synthesis of Recursive Functional Programs; a wait-free algorithm for weak refere...

microsoft.com
2
7
739
Load more