Joshua Levy
@ojoshe
Followers
4K
Following
16K
Media
456
Statuses
5K
AI, IA, writing, engineering, knowledge tools, truth, kindness, other arcana ▪︎ Author of guides with 2M+ readers ▪︎ Publisher @Holloway
SF
Joined December 2013
@BrianFeroldi @GRDecter The danger is not in letting students use tools like ChatGPT. It is in letting them learn that they don’t have to think, or believing that the tools are smarter than they are.
Now tools that create plausible yet dubious content are ubiquitous, good judgement and critical thinking are more important than ever. I’d like to see teachers help students critique AI-written essays. Find errors, describe what is incorrect or what you would write differently.
9
32
340
@RichDecibels If I were to give advice to a friend on depression, based on my own experience and that of others I've been close to, it would be this. (I tried to be brief but it's 3 pages.)
9
67
519
The area of emerging research around sunlight/infrared on COVID outcomes seems so significant, widely applicable, and low risk that far more people should know about it. I don't post about COVID much, but since I've seen little on Twitter about this, a 🧵 may be helpful. 1/13.
20
73
327
@AliAbdaal I spent 2+ years during the pandemic learning and experimenting with lots of bodyweight/calisthenics fitness programs. Many are great but few give a full picture. So I summarized it all in one chart. It's big. Still a draft. But it's the reference I wish I'd had starting out.
11
22
281
@alexalbert__ This is very cool but I have to ask. If it's so new and different, couldn't @AnthropicAI come up with a better name than "the new Claude 3.5 Sonnet"?. Like (just spitballing here) maybe "Claude 3.6"?.
11
3
226
GPT-4 passing the LSAT or GRE is incredibly impressive. At the same time I think we need a reminder of a logical fallacy we’ll see a lot of this week:. That software can pass a test designed for humans does not imply it has the same abilities as humans who pass the same test.
I don't give a damn about what is or isn't AGI. It doesn't matter. Below is GPT-4's performance on many standardized exams: BAR, LSAT, GRE, AP, etc. The truth is, GPT-4 can apply to Stanford as a student now. AI's reasoning ability is OFF THE CHARTS. Exponential growth is the
17
43
191
@dgardner It’s worth remembering it’s not websites that killed most local newspapers. It was killing revenue models, including bundled classifieds. And it wasn’t feed algorithms that caused Facebook to spread misinformation and divisiveness. It was their incentives to drive ad growth.
2
20
169
While packing books today, found one bit of tech history I’m proud to have inherited: The Internet, circa 1982. With the name and phone number of every. single. person.
4
46
172
@nickcammarata Salvational ideology is a hell of a drug. It would be fascinating to travel back in time a century or more to chat with idealistic political radicals (like Marxists during their early years) and compare it to today.
4
6
140
Coach and physiology prof @DrAndyGalpin recently gave an outstanding interview on @hubermanlab on how to assess your fitness across 8 dimensions. Essential for anyone who cares about their fitness. But there is a lot of info. I took notes, so here is a cheat sheet. 🧵
14
10
141
@AbstractFairy Most of the books you read are written by obsessive, outlier writers who call themselves “authors.”. More seriously, it’s the natural way media has always worked; it’s just that we have now added expanded the set of creators from a tiny fraction of a percent to a few percent.
1
2
138
@fireship_dev It seems GPT-4 is an amazing tool to read and apply developer docs to scaffold out code. But it’s missing real-world experience and pain points of executing, using, and maintaining that code. That’s the real value you get from explanations from senior engineers.
7
3
130
Thinking about the future today?. Predictions from 1922-23 for now (left) and from recently for the 2110s (right). One lesson: Tech predictions are uncannily accurate about half of the time—and completely wrong the other half. The tricky part is knowing which are which.
4
10
121
@Rainmaker1973 “You have a pint of strong beer that’s 10% alcohol. You want to add water to it so it’s 5% alcohol. How much water should you add?”. I think most people get the potatoes question wrong but get this one right. Perhaps intuitions about potatoes are worse than intuitions about beer.
4
1
126
The philosoher’s stone is a perfect analogy. The term “AGI” sounds scientific but the idea encapsulates so many assumptions. It makes AI discussions a psychological attractor for a broad range of hopes, fears, and fantasies about technology.
When I say I want to build "strong AI" or "general AI", I don't mean "AGI" in the sense that most everyone else means it. In its common use, "AGI" is a cultural construct akin to the Philosopher's Stone, that no one can define crisply but on which people project all kinds of.
2
19
72
Just sharing some beautiful old geometric studies of polyhedra from Max Brückner (who inspired MC Escher).
3
15
108
@northstardoll @Amina_io To dream big or to dream small costs the same. —Jorge Paulo Lemann.
1
12
104
@waitbutwhy Added one more layer on this diagram: how do these types of people do on social media?. The posts that get engagement are the ones at the extremes: they must make you feel good, feel smart, or feel angry. That’s why the most popular people online are the brave (but it helps to
8
9
105
@PakicetusAdapts @danielleboccell @nickcammarata A lot of depression is maladaptive helplessness. A loss of agency. The opposite of depression is not happiness; it’s vitality, which rests on agency. That comes from forward motion or the potential for change or growth. But also, growth requires pain. Experience is learning.
1
9
104
@danprimack Psychologically and statistically based on Trump’s past relationships, it’s a near certainty. I think the more important question is, what does this mean for the relationship between Tech in general and Washington.
3
1
97
Now tools that create plausible yet dubious content are ubiquitous, good judgement and critical thinking are more important than ever. I’d like to see teachers help students critique AI-written essays. Find errors, describe what is incorrect or what you would write differently.
@GRDecter Pandora’s box is open and there’s no going back. I completely understand the desire to ban it from schools, but it would be far more useful to teach kids how to best use it.
0
20
92
We’ve become big fans of @NotionHQ at @hollowayguides and thought we’d share some learnings. Anyone find these useful or have comments?
7
6
95
I've learned many interesting things about fitness lately. But one of the simplest and most useful is this:. There are essentially 9 ways the body adapts to exercise. Here they are. Summarized from @DrAndyGalpin and his incredible explainer videos (links below).
3
9
85
@s_r_constantin One on my mind recently and I’ve not seen stated this way (a kind of a loose analogy to Hanlon’s razor):. Don’t attribute to conspiracy what can be explained by systems of incentives.
1
5
84
A big question ahead is if and how watermarking of AI generated text plays out in practice. Scott Aaronson (OpenAI) explains how it works very nicely here, from his blog. Note it can be circumvented with enough effort, for example by using one AI to paraphrase another.
6
20
85
An excellent point. The more I use LLMs in practice the more it seems closely analogous to how years ago everyone (including students) had to learn how to use Wikipedia and search engines, when to trust them vs what’s SEO clickbait, etc.
When ChatGPT came out I thought I wouldn't use it for learning because of its tendency to slip in some BS among 10 helpful explanations. Then I tried it, and found that it forces me to think critically about every sentence, which is the most effective mindset for learning.
3
13
83
A great and authoritative summary of fat/carb/protein recommendations, especially for anyone physically active, packed into 5 minutes by @DrAndyGalpin:.
5
13
84
Okay, I generally think people overestimate AI systems, but this transcript mostly convinces me there is not just language structure but a personality inside Bing Chat. Based on Reddit context and posts from others I believe this is valid. Rest of convo and link below.
9
15
85
An important, thoughtful, and nuanced conversation with @GaryMarcus and @ezraklein on the current state of AI, ChatGPT, and the future. Do give it a listen. Here are 3 themes that I was glad were discussed. 🧵.
1
21
78
While I never worked there, old-time Google friends have told me one of the most valuable things @marissamayer did during her many years leading search experience was simple: saying no. For those early years Google said no to the death-by-a-thousand cuts changes to search.
2
9
76
@EcZachly I know analysts at major firms who make more than that and all they know is Outlook and Excel!. More seriously, another lesson here (for anyone, engineers included), is that having business objectives in context when you use any tool makes you far, far more valuable to a company.
1
4
82
@AlecStapp Especially notable:.- Highly effective .- Costs $1K+/month.- You gain the weight back if you stop.- It may be lifelong for many.- It’s now recommended for obese children.- Revenue of $900M in 2022, before the likely surge coming this year
6
0
69
We're launching the Open Guide to AWS today — basics, expert tips, and gotchas for @awscloud #opensource on @github
0
36
73
@AlphaSignalAI @Flawlessai This plus NVIDIA’s eyeball readjustments for zoom meetings makes me think we’re going to have to accept eyes and lips are frequently going to be digitally altered. We might all get hypersensitive to small glitches.
1
2
66
@elizalian It’s remarkable how so many disruptive and contrarian early adopters tend to think similarly all at once.
1
7
70
Covid rates are going up quickly. Two severely under-reported but essential pieces of COVID advice:. (1) If you have any hesitation about another booster vaccination, go get Novavax now it is available. It's a more traditional protein vaccine, data is good, and side effects much.
4
8
64
@dmvaldman It’s interesting to contemplate. Doug always was one of the rare voices pushing for IA (intelligence augmentation) over AI (smart software or robots). Yet even today, now we see how right he was about so much, so few really lean into his vision of making *people* smarter.
2
5
68
Tests for humans don’t test for the abilities most or all humans have. They test for the abilities that are difficult for humans.
2
11
68
@paulg @fearmerchant @Tawhidsquared @lexfridman Don’t worry if you’re a Twitter novice. Just remember 3 principles:. 1. Seek attention by any means necessary.2. Form opinions as quickly as possible.3. Take everything others say personally.
5
4
66
@bryan_caplan People are frequently confused or conflicted about this, but in fact there are five purposes of punishment:. 1. Deterrence (individual and general).2. Incapacitation (off the streets).3. Rehabilitation .4. Retribution.5. Restitution (paying damages).
4
3
63
@dhh There of course are institutional factors making it hard for them to ship products. But fundamentally, it’s also rational: the priority has always been to build free or cheap products (like Google Maps, Android, YouTube) that protect the enormously lucrative ad/search monopoly.
2
1
61
@mezaoptimizer More explained here. It seems, as he describes, that it will be possible to circumvent by using a different LLM to paraphrase the output of GPT. I expect this will be a possibly useful signal but not a reliable one.
A big question ahead is if and how watermarking of AI generated text plays out in practice. Scott Aaronson (OpenAI) explains how it works very nicely here, from his blog. Note it can be circumvented with enough effort, for example by using one AI to paraphrase another.
2
4
64
@HeidiPriebe1 Yes. Not just actions but words. If you truly believe you are unworthy and so not loved or important to others, then how could being critical of someone hurt them? In my experience, this especially applies not just to anxious but also to the fearful avoidant attachment style.
1
0
62
@dh7net It’s an interesting point. A movie can be excellent without a great soundtrack, yet for some movies a great soundtrack with top notch sound engineering makes it far better. It’s still the same movie but the stronger sensory experience enhances the power and immersion and meaning.
1
1
61
@oanaolt Just wanted to share a few thoughts on this from personal experience, as it may help others processing this. I’m sorry for what he went through and especially for those close to him. Losing anyone is terrible. But it’s uniquely difficult when it is to suicide or mental health.
1
0
57
@GergelyOrosz Mercurial was functionally quite close, solved the same biggest problems Git did relative to Subversion, was arguably easier to learn, and had some open source/Python momentum. My sense is it didn’t win simply because (1) Git was a bit more powerful and useful early on and (2).
7
1
52
The ironic thing is that by fearing we would create malevolent AI personas, we essentially already created them. In a digital world, our social consciousness is always written down. What we think lives on in training data, web searches, and models.
1
3
48
@skryl_alex @eastdakota @kevinroose One more key one: Microsoft has demonstrated the unthinkable: that a company of its size can purchase a great team and give it autonomy to grow. GitHub was not just strategically brilliant, it actually worked to scale and improve the product. Google simply doesn’t know how to.
2
3
53
@paulg @t_niemetz Similarly, this is purely speculative, but women are on average more agreeable as well, which could mean more amenable to taking a job unaligned with their degree.
1
2
53
@dsiroker @JayDanforth My alternate likely scenario:. 1. AI with powerful, impressive facet of intelligence develops.2. People amazed, predict AGI.3. People realize it’s superhuman at one thing and dumb at other things.4. New products, new economic shifts, get used to it.5. Return to 1 (N times).6. ?.
1
5
54
After publishing all-digital titles for 2 years now, it’s exciting to be able to share new books in print as well. This is a whole new toolchain on top of our own web publishing platform. It’s been fun and a challenge to map GitHub sources to high-quality typography on paper. .
We're thrilled to present our first ever print book. Angel Investing: Start to Finish is a the most comprehensive practical and legal handbook for angel investors and founders. And it looks real nice on a bookshelf. Get your copy here:
6
6
50
@WilliamAEden Excellent thread. I read it carefully. For what it’s worth, having followed and worked in the space a long time myself, I agree with almost all your conclusions, for both related and different reasons. It’s good to see a careful and moderated assessment amid hype and panic.
2
0
49
@anothercohen You don’t have a Google mindset. Next you’re going to tell me you don’t understand the value of having Google Meet, Hangouts Meet, Hangouts Chat, and Classic Hangouts be separate products.
1
0
52
I've talked to several engineers recently and it seems many still haven't figured out how exactly to unlock the power of GPT4. Here's a summary of my current approach with 9 specific suggestions. 🧵. 1. Think of GPT4 as three colleagues: an architect, a colleague, and an intern.
3
5
49
I think public economic discourse would be *so* much clearer if we all learned in school about the difference between rent-seeking vs wealth creation. If I charge $1000 to raise a chain I’ve put across a river, it’s not the same as starting a new riverboat service.
Who has a billion dollars? Your net worth is what someone is willing to pay for what you have, whether you've sold it or not. If a famous painter paints a new portrait, the value he's created wasn't taken from anyone. Nor if an entrepreneur creates a much-loved product.
4
0
49
@directxman12 One of the hard learned (and ever re-forgotten) lessons in software engineering is that wire formats and manually edited formats for structured data have different requirements. You wouldn’t want YAML as a wire format nor should you want JSON as a config format.
5
3
44
@LBacaj Usually we see posts with self-selected and oversimplified success stories, not the messy middle of trying new things and seeing them not go as planned. It’s kind of you to share this personal story. “Experience is what you get when you don’t get what you want.” —Randy Pausch.
1
1
46
@dmvaldman I don’t know what Engelbart would build as a demo today. But if I imagine showing him LLMs (I am thankful to have had a few conversations with him), I’m pretty sure he’d still be asking us how these tools can help humans be better at working together to solve important problems.
6
3
45
“The only way to handle danger is to face it. If you start getting frightened of it, then you make it worse. Because you project onto it all kinds of bogeys and threats that don’t exist in it at all. “Whenever you meet a ghost, don’t run away. Because the ghost will capture the.
“The only way to handle danger is to face it”—Alan Watts, 1958. How to discover our real ghosts. We face a new ghost and it is in a machine, they are calling it AGI. Meet the new ghost, same as the old ghost.
1
5
45
@StefanFSchubert Three things can be true at once:. (1) It can be quite hard to know the difference between a stupid question and a good question in the moment. (2) Given this, many people should bias toward asking what they fear to be stupid questions. (3) But for that to work, you must also.
1
2
44
@HeidiPriebe1 I found this really interesting too. The idea of a set point makes sense. But also I’ve never been convinced the sad-to-happy continuum is the right axis to focus on. Rather it’s the depressed-to-enlivened spectrum or perhaps low to high agency.
@PakicetusAdapts @danielleboccell @nickcammarata A lot of depression is maladaptive helplessness. A loss of agency. The opposite of depression is not happiness; it’s vitality, which rests on agency. That comes from forward motion or the potential for change or growth. But also, growth requires pain. Experience is learning.
3
1
44
Whoa in a sign of the times there is now a brain fog emoji but seems like no one noticed 😶🌫️.
1
2
37
@nickfloats Yeah. Any American who visits friends in Brazil should know this: always offer to hit the mall and buy brand name cosmetics and Apple products for them before you leave.
1
2
41
So how would you explain tech debt to junior engineers—or anyone who cares about engineers? . Here's my attempt, in one sketch. Good software engineering usually boils down to tough trade-offs. 🧵
3
6
39
@the_wilderless For some people, reading it at the right time is transformative. Others get annoyed by what they may to see as a pretentious or over-intellectual style and can’t get past that. Imo it’s a deeply thoughtful book with many worthwhile insights, if you are open to them. Example:.
“Science works with the chunks and bits and pieces of things with the continuity presumed, and [the artist] works only with the continuities of things and the chunks and bits and pieces presumed.”. —Robert Pirsig,. Zen and the Art of Motorcycle Maintenance. This is one of
0
1
39
@VicVijayakumar This is totally normal. But it’s friendlier to say “maintained.”. Try to avoid saying “cleaned up” or “finally fixed” or “rewrote it in OCaml and it’s so much faster now.”.
1
1
41
@utotranslucence It’s worth noting German fitness bread also comes in a sexier version. Nothing heats things up quite like pumpernickel.
5
2
40
@shiraeis @AlphaSignalAI @elonmusk Being “unbiased” seems like a hopeless goal since there will never be agreement or trust that it is. Ultimately, focusing on transparency (about training data and about guardrails) might be the most reasonable option.
2
2
39
@BrookeSellas Were you an only child? That might explain acquiring such essential knowledge late in life. Now I wonder, how many others have used a 9-volt battery to ignite steel wool? It’s quite spectacular (particularly if you’re 10 years old).
3
0
34
How on earth did no one ever tell me this. You can write select statements in Google sheets. 🤯
3
7
40
When do you know if you or someone else is good at a specific skill? A short life philosophy thread. It's a mental model I've found helpful: non-evident skills. Skills can come in three areas:. 1. Evident skills (objective).2. Non-evident skills (objective).3. Subjective skills.
3
1
34
@gothspiderbitch I hope this time you purchased whole nutmeg and a nutmeg grater. Whole nutmeg keeps well, so you’ll be good for another 30 years.
0
2
35
@deepfates I think the biggest issue with Spotify traces to one poor product choice:. Personalized recommendation playlists are great. But not all playlists should be personalized recommendations. If I ask for a radio based on a song, give me a radio based on the song. Not yet another.
3
2
37
@gbrl_dick The thing is, people seem to forget this, but no matter how smart the tools, they don’t eliminate the intrinsic complexity of problems, especially problems that involve humans (which are basically all problems).
1
3
40
@ESYudkowsky My take: This is nowhere near actual general intelligence, nor is it real anger. It’s likely way too much training data that includes confrontational conversations. But I think it is probably as dangerous as we fear, because of how it will be used, deliberately and accidentally.
3
2
37
@ESYudkowsky This question has some confusing implications. Even if we had perfect Turing and IQ tests, the shape of skills to pass the Turing test and the shape of skills to pass an IQ test are different. And neither is the same as the shape of the actual skills of a human with IQ of 100.
4
0
36
@40goingon28 After (2) comes the five stages of meteorological grief:. A. WTF.B. Nothing like this has ever happened before!.C. Oh, it has?.D. Mudslides? Damage? How is it no one planned for this? Who is responsible??.E. Look at those poor Minnesotans. God I’m glad we have such great weather.
0
1
29
If there’s one person to listen to now it’s the guy who (literally) wrote the book on pandemics. Michael Osterholm foresaw.- lack of preparedness.- aerosol transmission .- supply chain issues on PPE.- WHO+national leadership missed opptys . Please. We should listen now. Links👇.
1
8
37
So glad I bought this. A Grammar of Typography by Mark Argetsinger is just a stunning book, in its design and in its depth. It’s uniquely current on digital tools as well the history of traditional book design.
2
2
35
Since the days of Marvin Minsky and Doug Engelbart, there have been two broad visions: AI and IA. AI is what has always captured the public imagination. But Doug's IA vision (intelligence augmentation, as he called it), ever since his 1968 demo, has _actually_ shaped almost all.
I’m playing around with calling our tech, as it is today, IA (intelligence amplification) instead of AI. IA have the vibe of tools for thought, needing human interaction, and resemble a lot more what we actually have today. AI feels more like independent long-running agents.
4
5
34
@IntuitMachine This is interesting. But saying this as if the formality gives it weight is a bit silly. Excel, CSS, and awk are technically all Turing complete too. This statement doesn’t really tell you much at all about what something is useful for in practice.
3
1
37
@Empty_America This is a very American perspective. If you live a few places globally you realize air conditioning is a luxury in much of the world—and you notice the difference. You could say the same about indoor plumbing. It does offer some absolute, not just relative, benefit.
1
0
34
@dickiebush A mindset change around fitness, after learning that the minimum effective dose for resistance training is one set. Instead of aiming for more workouts, instead I just aim to reduce the friction to doing a workout of any size. Result for me was calisthenics 5+ days a week.
1
2
37