Rafael

3.2K posts

Rafael

Rafael

@rafa_akmr

building @1to1_data multi-modal human data for frontier ai

mia/sf Katılım Ocak 2017
714 Takip Edilen254 Takipçiler
Rafael retweetledi
Aaron Levie
Aaron Levie@levie·
CEOs are uniquely prone to AI psychosis because they’re sufficiently distant from the last mile of work that still has to happen to generate most value with AI. So when they play with AI, they see the happy path results, often not considering the next 10 or 20 things that have to happen to get sustainable results from agents. “Look I made this awesome product prototype”. Yes but you didn’t have to review the code before it went into production and fix a bunch of issues. “Look I generated a contract”. Yes but you didn’t verify all the terms before it goes out to the counterparty and didn’t have to wire up all the past contracts to work with. The best thing you can do as a CEO is to use AI a *ton* to figure out the real implications of agents in the enterprise, and come out the other side with an appreciation for both the upside and the real work that goes into them.
Michal Malewicz@michalmalewicz

CEOs are the most delusional about AI. Detached from reality.

English
207
430
4.1K
567.2K
Rafael
Rafael@rafa_akmr·
@itstuyo so basically deferred cashback presented as waived purchases same mechanism but with less control. cool launch video, questionable product
English
0
0
1
817
Tuyo
Tuyo@itstuyo·
We created a card that sometimes doesn't charge you. Buy Now, Pay Maybe.
English
758
237
6.3K
15.9M
Rafael retweetledi
Odyssey
Odyssey@odysseyml·
Why We Must Build World Models
Odyssey tweet media
English
14
40
351
21.3K
Rafael
Rafael@rafa_akmr·
@sporadica ppl are scrolling less bc they're using ai more
English
0
0
0
82
Rafael
Rafael@rafa_akmr·
44k hours and youre already seeing generalization to unseen envs if the curve holds, 1b+ hours is where world models start matching what text models did at 10t tokens
RoboPapers@RoboPapers

World models have many different uses, from evaluation to training data generation to robot planning. DreamDojo is a new foundation world model that allows for impressively general and long-horizon interaction, generating coherent videos for interaction sequences over a minute long. It works in a wide range of environments and even generalizes to previously-unseen environments. We talked to @ShenyuanGao and @willjhliang about how they built DreamDojo, and about what tricks were necessary to scale world model learning on data with sparse action labels, pretraining on 44,000 hours of human data and adapting to a wide variety of robots, environments, and skills. Watch Epsiode #77 of RoboPapers with @micoolcho and @chris_j_paxton now to learn more!

English
0
0
3
102
Rafael
Rafael@rafa_akmr·
if we design data that induces better representations, decision structure becomes more legible for cross-embodiment transfer that’s how we can actually understand the emergent capability we’re seeing from scale
Danfei Xu@danfei_xu

Gave a talk on Robot Learning from Human Data at Stanford. It was great to be back! Some opinionated points: 1. Human data collection capacity is outpacing the research. 2. We still don't have the "science" for scaling robot capability with human data. 3. We are far from being able to model naturalistic human behaviors. youtube.com/watch?v=NUtaN1…

English
0
0
2
121
Rafael
Rafael@rafa_akmr·
was great attending iclr in the motherland. couldnt ask for a better backdrop 🇧🇷 got to connect with folks training on the multimodal data we've been creating for months
Rafael tweet mediaRafael tweet mediaRafael tweet media
English
3
1
17
344
Rafael
Rafael@rafa_akmr·
closer to the vision @nikitabier but instead of static categories would be great if they could be dynamic/prompt-based x.com/rafa_akmr/stat…
Nikita Bier@nikitabier

Ladies and gentlemen, today we're launching one of our biggest changes to 𝕏 Introducing Custom Timelines This feature allows you to pin a specific topic to your home tab. With support for over 75 topics, you can dive deep into your favorite niche on X. It's powered by Grok's understanding of every post with the algorithm's personalization—meaning every timeline is made just for you. And it works even better when it's a topic you already engage with. This was a huge undertaking across many months, so we're excited for you take it for a spin. We're giving early access to Premium subscribers on iOS (and Android coming very soon).

English
0
0
0
103
Rafael retweetledi
orph
orph@orphcorp·
information is free because saturating your cognition & preventing you from taking action is the product
English
121
2.4K
15.6K
413.5K
Rafael retweetledi
Essam Sleiman
Essam Sleiman@essamsleiman·
tldr: everyone is converging on the same product shape: a general harness that takes a goal, uses tools, and does knowledge work. once every product is a harness, the next frontier is the feedback loop that improves it after deployment.
Nicholas Charriere@nichochar

x.com/i/article/2039…

English
37
66
1.1K
278.6K
Rafael retweetledi
Will McKelvey
Will McKelvey@Will_McKelvey·
The rise of recreational exercise over the last 50 years coincides pretty tightly with the US economy's shift from manual to intellectual labor. If AI automates a lot of intellectual labor, we'll see a similar rise in recreational education. Imagine taking an adults-only class on English lit: you would come away feeling intellectually stimulated, you'd sound more worldly at your next social engagement, and if the experience was in person, you might meet people with similar interests! This feels inevitable. People go to the gym to keep their bodies in shape for health + dating, so why wouldn't they take steps to keep their minds similarly sharp? Ultimately, this will look like a better-marketed version of community college adult-education classes. MasterClass was maybe just a few years too early...
English
22
8
114
9.4K
Rafael retweetledi
Startup Archive
Startup Archive@StartupArchive_·
Elad Gil: “Things that work tend to work pretty fast” “I do think there’s a bit of a myth in Silicon Valley that you should keep grinding no matter what and it’s just about perseverance, and I think that’s really bad advice… In general, things that work tend to work pretty fast and usually that’s within the first year of launch.” There are undoubtedly counterexamples, but Elad points out that when a startup finally works after five years of grinding, it’s usually because they changed direction. “When people pivot, they tend to pivot locally. They don’t tend to pivot across markets. And that’s a huge mistake because you get stuck in some local maxima… I think usually if the thing isn’t working and you decide to pivot, it’s often wise to rethink everything from the ground up. And it’s hard if you’ve raised a lot of money and have a big team.” To determine if their product is working, Elad urges founders to look for indicators that users care. It’s usually a sign that you’re onto something if you’re product is half broken and people are still using it (e.g. Twitter fail whale days) or if customers are constantly complaining but they keep sending you feature requests—at least they care enough to complain! Again, you’re looking for signs that people care. Video source: @southpkcommons (2023)
English
8
17
225
33.2K
Rafael
Rafael@rafa_akmr·
the future we thought we would get: flying cars the future we will get: financialized claude credits
t◎ny p@tonyplasencia3

@antoniogm I’ll pay for this guinness with my claude credits sir

English
0
0
1
83
Rafael
Rafael@rafa_akmr·
@PaulSkallas Yes and no. Something to be said about substack’s growth (if you consider that reading)
English
0
0
0
143
LindyMan
LindyMan@PaulSkallas·
"readers" don't really exist anymore. Reading books is a niche hobby like flying a kite or bird watching. So it attracts a certain type of person If you want a normal person to read your stuff you're going to have to get their attention while they are at work and bored
English
41
87
2.2K
74.1K
Rafael retweetledi
himanshu
himanshu@himanshustwts·
been closely looking into the emerging trends or paradigms lately where a good amount of cash + talent is flowing and now i am increasingly convinced that RL environments and Data for Robotics/Simulation is going to have generational run.
English
13
7
247
28.6K
Rafael retweetledi
signüll
signüll@signulll·
at some point ppl will have a token burn amount in their comp packages (maybe as a perk). $250k base $5m equity $250k year token burn seems quite inevitable at least in the short term while humans are overseeing ai.
English
54
19
751
56.3K
Rafael
Rafael@rafa_akmr·
Evaluating staff on token consumption and outcomes per token💡
Aaron Levie@levie

Without getting into the specific numbers, this underlying concept and trend is going to be very real. For any worker who is able to wield AI agents effectively in an organization, their compute budgets are just going to monotonically go up over time. This will of course start in engineering, where we already know developers can run multiple agents in parallel, or have projects going over night. But this eventually hit the rest of knowledge work as well. Lawyers that can create and review more drafts, marketed that can build more campaigns and test more ideals in parallel, sales reps that can reach out to more customers and process more leads. Many of these activities will essentially be token-dependent in how much work a single person can do. These aren’t chatbot workflows answering a simple question, but agents that are running and processing through incredible amounts of data at scale, and generating all new forms of information. Companies will have to figure out how they budget for this, and it likely won’t be an IT budget item over time, but ultimately owned and allocated by the business. Maybe the CFO is ultimately the head of AI :-).

English
0
0
0
60
Rafael
Rafael@rafa_akmr·
Do you ever wake up and think like damn we really have superintelligence in a piece of silicon that can satiate infinite intellectual curiosity
English
1
0
1
67

Keşfet

@itstuyo @sporadica @horwitzben @AxelMtbr @nikitabier @southpkcommons @PaulSkallas @elonmusk