Judah Newman

0 followers · 858 views

on the atlas — 31

highlights — 178

Some additional notes: - we can freely choose the dimension of the latent space to suit the problem at hand: small latent spaces for simple problems and larger latent spaces for problems with complex dynamics
Deep Markov Model — Pyro Tutorials 1.8.4 documentation
There are now six zillion “AI tutor” startups in flight, but none I’ve seen have yet felt like they’re on the right track. I don’t have a clear sense of what my complaint is: too instructional, too interested in “right answers”, not interested enough in discussion and sense-making.
Memory systems and problem-solving practice | Patreon
I think, however, that there isn’t any solution to this problem of education other than to realize that the best teaching can be done only when there is a direct individual relationship between a student and a good teacher—a situation in which the student discusses the ideas, thinks about the things, and talks about the things. It’s impossible to learn very much by simply sitting in a lecture, or even by simply doing problems that are assigned.
Memory systems and problem-solving practice | Patreon
This is an important distinction to understand, and I fear it’s one that memory system designers consistently ignore: the appropriate intervention for a “wrong” answer will vary enormously with the nature of the prompt.
Memory systems and problem-solving practice | Patreon
f you want both flexibility and durable fluency, then you should practice, but along a gentle slope. You don't want to overload your working memory so much that your mind can’t reorganize its representations of the relevant concepts.
Memory systems and problem-solving practice | Patreon
The authors suggest that this experiment’s practice problems were varied enough to generalize over, and the cognitive load was light enough that schema acquisition was possible.
Memory systems and problem-solving practice | Patreon
The literature calls this “schema induction”. The claim is that when experts solve problems, they lean heavily on schemas, structures “that permit problem solvers to categorize a problem as one which allows certain moves for solution”. To build flexible problem-solving capacity, you need to acquire these schemas. That’s often done through induction—noticing patterns across a set of problems, noticing that a particular set of moves seems to help.
Memory systems and problem-solving practice | Patreon
O’Day’s experiments suggest that to become good at solving problems, you need to practice solving problems, ideally with feedback.
Memory systems and problem-solving practice | Patreon
It's not necessarily enough to just practice recalling some information: practice should require processing that information in the same way you expect to be processing it when you want to use it later.
Memory systems and problem-solving practice | Patreon
The ultimate goal of CBT is to challenge and modify negative automatic thinking patterns that maintain a person’s emotional symptoms.
Cognitive Behaviour Therapy: A Cybernetic Perspective
CAs involve a goal, strategies to achieve this goal, and an ongoing process of feedback and interpretation of information to assess whether progress is being made.
Cybernetics: The Key to a Unifying Framework for Psychology
One researcher I talked to said the arguments for acceleration made sense five years ago, when there was almost nothing worth experimenting on, but that they no longer think this is true.
OpenAI's "Planning For AGI And Beyond" - by Scott Alexander
The alignment researchers I’ve talked to say they’ve already got their hands full with existing AIs. Probably they could do better work with more advanced models, but it’s not an overwhelming factor, and they would be happiest getting to really understand what’s going on now before the next generation comes out.
OpenAI's "Planning For AGI And Beyond" - by Scott Alexander
DeepMind thought they were establishing a lead in 2008, but OpenAI has caught up to them. OpenAI thought they were establishing a lead the past two years, but a few months after they came out with GPT, at least Google, Facebook, and Anthropic had comparable large language models; a few months after they came out with DALL-E, random nobody startups came out with StableDiffusion and MidJourney. None of this research has established a commanding lead, it’s just moved everyone forward together and burned timelines for no reason.
OpenAI's "Planning For AGI And Beyond" - by Scott Alexander
So we want to maximize the amount of time it takes between IQ 200 AIs and IQ 1000 AIs.
OpenAI's "Planning For AGI And Beyond" - by Scott Alexander
But if they burn their lead in 5-10 years, right before the dangerous AIs appear, the alignment researchers will have two extra years studying how to prevent advanced AGIs from making superweapons, which is more valuable.
OpenAI's "Planning For AGI And Beyond" - by Scott Alexander
he prompt-writing process may truly need the feedback pressure from subsequent review sessions, to shape the prompts appropriately.
Becoming a Wizard-of-Oz learning assistant | Andy Matuschak on Patreon
Worse: it’s usually not at all straightforward to evaluate whether the prompt is on-target, nor to articulate the way in which it’s off-target.
Becoming a Wizard-of-Oz learning assistant | Andy Matuschak on Patreon
Often I’d written a prompt that used all the right key terms, so it seemed right initially—but when Alex practices it, he notices that it’s not reinforcing quite the right aspect. Or I’ve written a prompt about something that had confused him, and now he can answer it… but he’s still confused; there’s something important I failed to capture
Becoming a Wizard-of-Oz learning assistant | Andy Matuschak on Patreon
So the way you train your algorithm ends up forcing people to create a certain type of content, and what we have to be very careful with, I think on this particular point, is to not reward behavior where you obfuscate the actual content of the article, aka clickbait, where people write more sensationalist headlines or more sensationalist articles.
An Interview with Kevin Systrom and Mike Krieger About Artifact – Stratechery by Ben Thompson
One of the points he had is that the way these algorithms are trained, the objective function that you give these algorithms to optimize, ends up having this strange, almost tertiary effect that it forces content creators to start creating content that fits what the algorithm wants to optimize for.
An Interview with Kevin Systrom and Mike Krieger About Artifact – Stratechery by Ben Thompson
Right, because it’s like cutting out the middleman, right? Instead of trying to collect a bunch of almost right articles that are in your interest, wouldn’t it be great to pull everything out of them and just give you the one-page summary of what happened in F1 yesterday or whatever it might be?
An Interview with Kevin Systrom and Mike Krieger About Artifact – Stratechery by Ben Thompson
Yeah, it’s all good. I will tell you from my perspective, one of the wonderful things about having sold our company is actually that you can leave it.
An Interview with Kevin Systrom and Mike Krieger About Artifact – Stratechery by Ben Thompson
But you have to change what you’re doing drastically almost every three years, to stay alive
An Interview with Kevin Systrom and Mike Krieger About Artifact – Stratechery by Ben Thompson
“Wow, every time I see a startup stick to their original, what I’ll call their sacred cow or their 140 characters, that never works out in the long run.” Our version was square images, our version was just photos, our version was Feed.
An Interview with Kevin Systrom and Mike Krieger About Artifact – Stratechery by Ben Thompson
Our aspiration eventually is to let anyone be a publisher and one of the reasons why you can’t be a publisher right now is because you can’t find an audience. If you’re lucky enough to find an audience, then you go on Substack. If you’re lucky enough to find an audience, maybe you graduate to your own hosted blog or maybe you get hired by one of these big publishing groups, but it’s really hard to find an audience.
An Interview with Kevin Systrom and Mike Krieger About Artifact – Stratechery by Ben Thompson
I didn’t realize Chris wrote that post, but I should maybe in the back of my mind that’s what I was quoting and maybe I had read it, but I think all social networks start off as a utility and blossom into social networks long term, and that’s because they earn the right to be a social network, not because they start as a social network.
An Interview with Kevin Systrom and Mike Krieger About Artifact – Stratechery by Ben Thompson
Maybe we have some new announcement where they’ll bring it back in and I think as long as people give us enough tries over time I’m confident we keep getting better.
An Interview with Kevin Systrom and Mike Krieger About Artifact – Stratechery by Ben Thompson
Over the last couple weeks I’m like, “Wow, I’m actually finding stuff that I wouldn’t have found otherwise”, which obviously is the value proposition.
An Interview with Kevin Systrom and Mike Krieger About Artifact – Stratechery by Ben Thompson
You remembered the pain of only being two people and maybe forgot some of the benefits.
An Interview with Kevin Systrom and Mike Krieger About Artifact – Stratechery by Ben Thompson
But I mean, I feel like I speak for both Mike and me in saying that it never gets easier the second time around, but you know what to expect.
An Interview with Kevin Systrom and Mike Krieger About Artifact – Stratechery by Ben Thompson
I say to Mike, you can go to all these bookstores or these conferences and get advice about what to do the first time and what not to do the first time, but there’s literally zero books or zero conferences that focus on what not to do the second time and it turns out you just make a whole host of different mistakes.
An Interview with Kevin Systrom and Mike Krieger About Artifact – Stratechery by Ben Thompson
Everyone assumes Apple TV shows are good. It reminds me there’s that argument I think Peter Thiel made in Zero to One that it behooves a monopoly to get into businesses where they lose money that are very visible because that detracts attention from other places.
An Interview with Matthew Ball About Disney, Streaming, and the Metaverse – Stratechery by Ben Thompson
there’s no provable law here, but it’s generally my experience that the people with the best advice and relevant experience are too busy doing things to be consistently hanging out in large communities.
Building your executive network. | Irrational Exuberance
However, I generally believe that creating content is a time-inefficient way to build your network. If you’re excited to write and speak, then absolutely do that, it will help you build your network, but if you just want to build connections, then I strongly recommend pursuing other avenues.
Building your executive network. | Irrational Exuberance
By far the best way to build your network is to spend time working at a large, fast growing company in a central tech hub like San Francisco, New York or London. Five years later, your colleagues will have spread across the industry. It’s a bit like meeting friends in college: it just happens without much deliberate effort. I highly encourage folks pursuing executive roles to spend at least a year or two in this sort of company early-ish in their career. It’s not the only way to build out your network, but it is by far the most cost effective.
Building your executive network. | Irrational Exuberance
This is difficult, particularly as a newer executive, because you’re likely ahead of your peer group in your career. Even if you know hundreds of folks in the industry, only a few of them may have experience as an engineering executive.
Building your executive network. | Irrational Exuberance
Paid networks monetize by growing their attendees, which compromises their ability to prioritize high-quality participants.
Building your executive network. | Irrational Exuberance
In most of my roles, I’ve learned more from my peers than from my manager. Even when you get along well with your manager, your peers’ perspective will usually be closer to yours than your manager’s
Building your executive network. | Irrational Exuberance
Importantly, I think that the factors themselves are best developed in that zone. So it may be futile to ask someone who is panicking to focus on tranquility IF they have not focused on tranquility before when they were in the optimal zone.
Buddhist Annealing: Wireheading Done Right with the Seven Factors of Awakening | Qualia Computing
In this light, when your mind is too excited and restless you should put emphasis on the tranquilizing factors (concentration, equanimity, and tranquility). When your mind is too sluggish or sleepy, you put extra emphasis on the exciting factors (investigation, energy, and joy). And the balancing factor (mindfulness) is what allows you to keep track of this process without getting distracted by the hindrances.
Buddhist Annealing: Wireheading Done Right with the Seven Factors of Awakening | Qualia Computing
Thus, while a lot of the meditative path is nominally about “renouncing” the pursuit of high-valence states of consciousness (and thus avoid the pleasure paradox), the result is nonetheless a state of consciousness that is high-valence in nature! Paradoxical? I don’t think so. The confusion is merely the result of conflating recipe and review
The Supreme State of Unconsciousness: Classical Enlightenment from the Point of View of Valence Structuralism | Qualia Computing
More so, over time, you entrain (and rewire!) the nervous system to become highly efficient at stress-dissipation. Dissonance is still there, but it “unfolds” and gets “metabolized” so fast that it barely counts as suffering. Highly annealed nervous systems are powerful stress-dissipation engines!
The Supreme State of Unconsciousness: Classical Enlightenment from the Point of View of Valence Structuralism | Qualia Computing
While Culadasa seems to emphasize the joyous journey and adventure of life, which may speak to an opposite weighting in favor of the positive valence being worth the negative valence that comes along with it. Certainly not all spiritual roads lead to Rome.
The Supreme State of Unconsciousness: Classical Enlightenment from the Point of View of Valence Structuralism | Qualia Computing
In contrast, it turns out that there are a manifold of ways of experiencing excitement about the future in an epistemologically clear way and no delusions.
The Supreme State of Unconsciousness: Classical Enlightenment from the Point of View of Valence Structuralism | Qualia Computing
altitude, ask if each of your guiding policies is applicable, enforced, and creates leverage:
Writing an engineering strategy. | Irrational Exuberance
Avoid this sort of internal competition by ensuring your engineering strategy clearly articulates resourcing and priorities.
Writing an engineering strategy. | Irrational Exuberance
While it’s reasonable to be wary of rolling out strategy too early, every bad strategy rollout I’ve seen has been rooted in the executive’s inability to listen to clear feedback. Waiting wouldn’t have improved their strategy anyway.
Writing an engineering strategy. | Irrational Exuberance
I argued that writing an engineering strategies is like being a historian: look at how things are already working, write them down, and share what you’ve written:
Writing an engineering strategy. | Irrational Exuberance
A strategy is composed of three parts: Diagnosis: a theory describing the nature of the challenge. This is trying to identify the root cause(s) at play, for example, “high work-in-progress is preventing us from finishing any tasks, so we are increasingly behind each sprint” might be a good diagnosis Guiding policies: the approaches you’ll apply to grapple with the challenge. Guiding policies are typically going to be implicit or explicit tradeoffs. For example, a guiding policy might be “only hire for most urgent team, do not spread hires across all teams.” If a guiding policy doesn’t imply a …
Writing an engineering strategy. | Irrational Exuberance