Growth strategy, emerging technology, management innovation, analytics, and business launch. “Luck Favors the Prepared Mind" #learn #energize #achieve

Washington, DC
Joined March 2009
“Civilization advances by extending the number of important operations which we can perform without thinking about them." Alfred North Whitehead
1
6
Leif Ulstrup retweeted
Replying to @curious_vii
was begging for a matrix 😂
1
1
1
3
373
Leif Ulstrup retweeted
Meta spent as much as the Manhattan Project on GPUs in inflation-adjusted dollars. Really curious how long they will continue to release open models with that sort of capital expenditure and models that are getting more economically valuable.
Training Llama 4 on 100k H100s.
22
101
12
707
91,431
How to Prompt NotebookLM's Interests One thing we hear constantly from users first experiencing Audio Overviews is how good the hosts are at uncovering the interesting bits from their sources. You can elicit that same interest-driven summarization in text chat too. Here's how: The Gemini models are amazingly deft at exploring large bodies of text and imagery to find specific sections that are high in surprise, high in unexpected information. You can call that just another incremental improvement on search and summarization if you want—to me it seems like a bigger deal than that—but either way, it lets you get answers to questions that no computer in the world could have produced just a year ago. When I'm trying to get my bearings with a few new documents that I think I need to understand, I'll load them into Notebook and ask a variation on: What are the most surprising or interesting pieces of information or narratives in these sources? And I'll maybe give it a gentle steer: Please focus on the NASA astronauts of the 1960s, not the later ones. And I'll tell it to include key quotes. In 30 seconds or less I'll have an enormously useful text document highlighting the most interesting and compelling passages in my sources. You can get most of this in NotebookLM right now, just by choosing to convert your sources into a Briefing Doc in the Notebook Guide panel. (And obviously, if you want to listen to this information in conversation form, Audio Overviews has you covered.) Both Briefing Doc and Audio Overviews are designed explicitly to surface interesting material. But you can get more clever and more personalized with it just by tweaking your prompts slightly. Here's one one example. I uploaded something like 500K words of transcripts from the NASA oral history project, covering the entire span of NASA from Gemini (the other one) to Apollo and all the way to the Space Shuttle. And then I asked NotebookLM: I'm interested in writing something about the Apollo 1 fire. What are the most surprising facts or ideas related to the fire discussed in these transcripts. Include key quotes. Take a look at the answer that NotebookLM generated in 20 seconds or so. How long would it have taken me to assemble this document manually, sifting through effectively five books worth of transcripts? 10 hours? You can't command-F for "interesting things."
28
99
18
767
73,556
Here’s a test to see how susceptible you are to conspiratorial thinking (it's fun) and a handy link to send to someone who you think could benefit from a critical thinking refresher course (again, it's fun, they won't be offended): theconspiracytest.org
1
6
1
23
3,454
Leif Ulstrup retweeted
The Homework Apocalypse already happened. AI can do basically any assignment and is being used everywhere. We need to start grappling with what that means, and how to use the capabilities of AI to get students to think, rather than replacing thinking. oneusefulthing.org/p/post-ap…
63
141
43
756
119,134
Programming is changing so fast... I'm trying VS Code Cursor + Sonnet 3.5 instead of GitHub Copilot again and I think it's now a net win. Just empirically, over the last few days most of my "programming" is now writing English (prompting and then reviewing and editing the generated diffs), and doing a bit of "half-coding" where you write the first chunk of the code you'd like, maybe comment it a bit so the LLM knows what the plan is, and then tab tab tab through completions. Sometimes you get a 100-line diff to your code that nails it, which could have taken 10+ minutes before. I still don't think I got sufficiently used to all the features. It's a bit like learning to code all over again but I basically can't imagine going back to "unassisted" coding at this point, which was the only possibility just ~3 years ago.
511
1,987
634
18,013
2,832,278
AI, at its best, doesn't give you answers; it helps you ask better questions. (These models aren't search engines.) They're more like microscopes or telescopes, making parts of information space legible that you otherwise might have missed. Which bits ultimately matter, where you invest your limited attention, the right course of action ... these are all up to YOU to figure out. Which is exciting — because as we're afforded more degrees of freedom, not just in terms of the "how" or the "what," but also the greater depths of "why" that we can investigate with significantly fewer resources. More whys at lower cost -> new frontiers opening up for everyone (in a business context, that means more product experiments at lower cost, meeting previously inaccessible customer needs, and more time for cross-functional creation, relation, and contemplation) and all that makes me a lot more enthusiastic about the future!
1
6
224
Leif Ulstrup retweeted
The best new ideas often come from taking something that’s traditionally niche or expensive, and making it broadly accessible. In AI, the biggest opportunities will be from taking work that most businesses can’t afford or don’t have access to, and making it broadly accessible.
25
76
15
612
151,996
Leif Ulstrup retweeted
Happy birthday to #BASIC, the programming language launched at @Dartmouth #otd in 1964 to encourage non-STEM students to use computers: bit.ly/1k89tw4 (v/@TIME)
9
157
16
426
43,334
Leif Ulstrup retweeted
At a conference at MIT, talking with a range of CEOs, technical leads in firms, etc. These are mostly people ahead of the curve in experimenting with AI in their organization. Every one of the ones applying it was saying it had a big impact on performance in their organization.
11
43
5
363
54,379
Leif Ulstrup retweeted
As a reminder, GPT-4 class models out-innovate and out-persuade the average human. An open sourced model of that power is going to lead to lots of unanticipated effects, good and bad.
Meta released their open source AI, Llama 3, today. As a key leader in LLMs, their models are often the most advanced open source ones out there. Based on benchmarks, the current model is not quite GPT-4 class, but their larger one (still training) will reach GPT-4 level.
8
35
3
227
29,804
Leif Ulstrup retweeted
One problem is the usual pathways for working with technology don't work here. The consultants don't have special knowledge you don't & you need to use the systems to understand what AI can do for you. Otherwise, it is all just talk & feels like fiction. oneusefulthing.org/p/strateg…
4
18
146
20,435
UCLA has recorded nearly 12 inches of rain in the last 24 hours, a 1 in 1000-year rainfall rainfall event for Westwood. A truly unprecedented storm in modern history for the region as an atmospheric river stalls over the region.
254
1,770
289
9,135
2,343,998
4-step process I use to build artisanal software (or deliver AI-powered SOPs):
reducing time-to-capability is one of the core benefits of AI really interesting things happen when you reduce the time it takes to get a result by 90% and then, scale up quickly with artisanal software (i.e., LLM-powered automation) so you can 10x output practically, you'll be best able to take advantage of AI if you cultivate a mindset where you set significantly more ambitious goals + apply extreme constraints this forces more fundamental changes in thinking, planning, and doing for individuals and organizations (a "re-architecture") ... insane amount of opportunity on the other side of such a change
2
1
6
658
New YouTube video: 1hr general-audience introduction to Large Language Models youtube.com/watch?v=zjkBMFhN… Based on a 30min talk I gave recently; It tries to be non-technical intro, covers mental models for LLM inference, training, finetuning, the emerging LLM OS and LLM Security.
534
2,723
609
16,464
5,128,417
One can think of AI as: ⚡️electricity (big deal, in everything, utility business model) ✍️ assistant (naive & eager-to-please analyst, copilot, or scribe) 🧠 thought partner (w/ extremely customizable information surface area and Q&A scripting) Third is most interesting imo
1
2
499
The billion dollar question is, "How do I create a machine that continuously acquires high-quality data on the frontier, gaining depth in a particular direction that exceeds the rate of expansion of what's 'in sample' for foundation models?" We can assume that all information and knowledge is eventually commoditized AND that profit can be found in the space between data-generating contact with a new pocket of reality and that data entering the GPT-X pre-training set. There are and will continue to be short-lived arbitrage opportunities to index and pre-package information that's already been digitized before it's turned into a commodity. What's more interesting is the problem of doing this in a way that's coherent and durable — by finding real-world problems, capturing insights by solving them, and then reinvesting those insights into accelerating forward movement in that problem space. Gotta think like an architect, not a house flipper.
1
1
3
1,260
Excited to share @GoogleDeepMind’s newest AI model GraphCast: the most accurate 10-day global weather forecasting system in the world. GraphCast can also offer earlier warnings of extreme weather events, including the path of hurricanes. In @Science today dpmd.ai/graphcast
107
753
117
4,044
680,581
Last week, @POTUS issued an Executive Order on AI. Already, federal agencies have identified >700 use cases, from "Autonomous Situational Awareness" for @DHSgov to "Smart Wildlife Tracking" for @Interior. Learn how USG is starting to think about AI: chat.openai.com/g/g-6FILh4tz…
1
1
5
431
Leif Ulstrup retweeted
A leading scholar in the study of working from home summarizes the research. WfH rates have stabilized, and the productivity impact of hybrid work (not fully remote work) is right around zero. It is likely to be a permanent part of work, and we are figuring out how to do it well
Our paper the "Evolution of Working from Home" just published pubs.aeaweb.org/doi/pdfplus/… overviewing the latest WFH data, trends, and research. Summary points: 1) WFH levels dropped in 2020-2022, then stabilized in 2023 2) Self-employed and gig workers are 3x more likely to be fully remote than salary workers (if you are your own boss you WFH a lot more) 3) Huge variation by industry, with IT having 5x WFH level of food service 4) WFH rises with density, and is 2x higher in cities than rural areas 5) WFH levels peak for folks in their 30s and early 40s (kids at home), those in their 20s have lower levels (mentoring, socializing and small living spaces) 6) Similar WFH levels by gender pre, during and post-pandemic 7) Much higher levels of WFH for graduates with kids under 14 at home 8) Productivity impact of hybrid WFH about zero. Productivity impact of fully-remote varied, dependent on how well managed this is. 9) Future will see rising levels of fully remote (the Nike Swoosh).
1
29
2
131
28,839