Former @WhiteHouse Tech Team, @CMSGov Sr. Research Scientist. Building Enterprise Data Science (#AI, #ML, #NLP). Boston Children's #DigitalHealth Advisory Board
Boston, MA
Joined June 2009
- Tweets10.4K
- Following1.8K
- Followers7K
- Likes19.9K
Aman Bhandari retweeted
The conclusion that @sayashk and I have reached over and over: direct harms from gen AI (that you can regulate) pale in comparison to the cost of societal adaptation and of the extractive business model. Without structural reform, those will continue. aisnakeoil.com/p/ai-safety-i…
ALT Unfortunately, downstream defenses against misuse impose a great cost on the rest of society. For example, the fact that any image or video could be AI-generated means that realism is no longer a marker of authenticity. That means we all need to adapt and change how we assess the veracity of information online. And that’s just one of dozens of such adaptations needed. Morally speaking, developers should bear some of the societal costs of harmful uses of AI, mirroring the fact that they reap profits from beneficial uses of AI. But legally speaking, we have no tool to enforce that. Remedying this situation is the great challenge of AI policy — a point we’ve made over and over. No amount of “guardrails” will close this gap.
Aman Bhandari retweeted
Why the rush to use emerging, experimental AI technology in high-stakes settings? As Emily Tavoulareas writes, people and organizations are integrating LLMs and chatbots into their lives and work as if the technology is reliable — but it is clearly not. techpolicy.press/sure-no-one…
Aman Bhandari retweeted
This is true, and why AI doing the writing for you is intrinsically limited. The act of writing is the act of *thinking*.
If you're not doing the writing yourself, the end product will be superficial and banal because the prompts were underthought
“You don’t imagine something first and then write it down. It’s through the act of writing that ideas make themselves known.” nautil.us/the-magic-of-the-b…
Aman Bhandari retweeted
“You don’t imagine something first and then write it down. It’s through the act of writing that ideas make themselves known.” nautil.us/the-magic-of-the-b…
Aman Bhandari retweeted
Impressive timeline as @DanaFarber governance addressed needs and constraints to allow for rapid deployment of generative AI at their institution. @NEJM_AI ai.nejm.org/doi/full/10.1056… Example for other research institutions and academic centers. #cancer #healthcare #biomedicalResearch
Aman Bhandari retweeted
This is a useful thread of recent papers showing that LLMs are biased when judging people in HR contexts. The papers offer some potential solutions, but it is important to recognize that current systems carry subtle biases that can be hard to recognize without extensive testing.
Aman Bhandari retweeted
I have talked to a lot of executives buying Microsoft Copilot for their firms. Not one seems to have considered what it means to suddenly automate the vast majority of management writing without training or reconsidering the meaning of the work.
From my book, Co-Intelligence:
Aman Bhandari retweeted
Scope is the enemy of proactivity on data teams. Proactivity is all about seeing what's happening and understanding what's coming next so you can be ready for it. The more things your team is responsible for, the more difficult this becomes.
Aman Bhandari retweeted
I work on health-inspired AI, and often find it difficult to be taken seriously in the ML community (e.g., "you should submit to a more specialized venue"). This paper sums up my experience perfectly + highlights how beneficial application-driven ML can be to the field!
What drives innovation in machine learning?
In a new paper, we argue that application-driven work is systemically under-valued in the machine learning community, but that it's essential for both innovation and impact.
arxiv.org/abs/2403.17381
1/3
ALT A figure with two columns. Under "Methods-driven ML" are the attributes "stereotyped benchmarks", "narrow set of evaluation metrics", "massive datasets", and "problem-agnostic methods". Under "Application-driven-ML" are the attributes "real-world tasks", "application-specific evaluation metrics", "auxiliary domain knowledge", and "use-inspired methods".
Aman Bhandari retweeted
Noticing an interesting trend in the world of techbio, where people with no research background say ‘when I worked in lab’ and they mean their one semester in a bio class. Never in my wildest dreams did i think the daily slog I did for 10+ yrs would become a desirable trait 😂
Aman Bhandari retweeted
More from @LifeSciVC on AI: I think true generative AI—coming up with brand new molecules never before seen by man, I think that remains elusive. A lot of the discovery-focused AI shops are mining patent literature and adding methyl groups here and there to existing things—and that has value. But separating that from the sort of hype of, hey, we're going to turn a four year discovery process into four days & a computer and it's going to be great and we're going to be cranking out tons of drugs—I think that's just not borne out by history. Right? What was it the cover of Fortune magazine in 1981 was Merck’s new computer that was going to design drugs? And here we are, four decades later, and you know, we're still learning how to apply all this powerful computation to making new medicines. I'm a big fan of it. We put money where our mouth is on that. Starting Nimbus back in 2009 as a computer-focused drug company this, and every one of our companies has some version of that in its wheelhouse around AI and machine learning—big compute, Big Data, computer processing. I’m enthusiastic about it, you can't dismiss it. But I think a level of cautious optimism is required.
Great conversation between Simone/@biocentury & Bruce/@atlasventure. On impact of AI on drug discovery/development, @LifeSciVC said: I think anyone today has to acknowledge the power of AI & machine learning to solve specific problems here. I do worry how far hype has gone beyond reality. There are clearly great opportunities to take large datasets— whether it's patient datasets, clinical datasets, and the like and apply AI & machine learning to them to identify patterns in there that might pick targets that might help you identify which patients are likely to respond.
Aman Bhandari retweeted
How do healthcare systems considering AI avoid the pitfalls articulated by Participant 7 below? @NEJM_AI ai.nejm.org/doi/full/10.1056… #practicalAIGovernance
Aman Bhandari retweeted
How can we build an economy enabling all people to have what they need to experience wellbeing?
Excited to announce a new Health & Political Economy Project to tackle such questions. @DarrickHamilton @victorroy
Join us for our 4/2 launch event in NYC! bit.ly/newagendahealth
Aman Bhandari retweeted
📢 #CallforPapers! JMIR Medical Informatics seeks contributions for a new section on "AI Language Models in Health Care." Explore real-world impacts, challenges, & strategies.
Learn More: jmirpublications.com/announc……
#aiinhealthcare #medicalinformatics
Aman Bhandari retweeted
Two things can be true at the same time:
1. Pretty much no one anticipated what could be done in 2023 with next-token prediction models.
2. The notion that current LLMs possess human-level language understanding is delusional.
Aman Bhandari retweeted
Accenture is doing more Gen AI revenue than all VC backed startups combined
Accenture’s GenAI bookings run rate grew to $2.4B
GenAI bookings grew 33% QoQ, or 313% YoY when annualized
GenAI bookings represent ~3% of total bookings for Accenture
Aman Bhandari retweeted
For years I’ve been interviewing data annotation workers who are the lifeblood of the AI industry. For years I’ve heard the same story: the platforms they work for wield total power, leaving them precarious & vulnerable to exploitation. A horrible example of this just happened 1/