turkeygizzard

I was going to say, AI has really enabled my hardware projects recently. I've found incorrect output is less common with the newest models - and they're capable of drawing decent diagrams too which can really help me sanity check some of their ideas. I've also had some success with asking it to cite sources where possible, so I get lots of useful links into robotics forums like chief delphi and similar which also help sure up the math

The Physiology of Climbing. If it looks interesting to you let me know and I can send you a copy!

The Physiology of Climbing. I can mail you a free copy once I get back home this weekend - I’d love any feedback!

To be clear, I aimed to avoid prescribing certain routines through most of the book. I wanted to basically provide a knowledge foundation for readers to evaluate routines or create their own. So instead of saying eg you should campus board, I try to explain that power has to be trained separately from max strength if you care about increasing power

Shameless self promotion but this is exactly how I ended up writing a book on strength training for climbing, just pursuing every rabbit hole I could.

I was ready to self publish but found a publisher who was interested. I had to make some changes to make it more readable, but you might have luck approaching publishers yourself

The title seems a bit hyperbolic compared to the article. It does briefly mention cardiovascular risk, but I was able to immediately find a meta-analysis showing no correlation.

If the gripe is with processed foods containing protein, then sure maybe there's a risk compensation argument, but personally speaking I buy Halo Top when I'm craving ice cream, not as a way to avoid eating chicken.

I also imagine that the target audience for these products are people who are relatively active and in that case the ideal protein consumption numbers are generally accepted to be significantly higher than the 0.8g/kg cited in the article.

Really appreciate you sharing your perspective. I recently wrote a book as a passion project and have been sitting anxiously on a contract. I'm not concerned about the money (I don't think my book will be a huge thing). My main motivation for going trad is the credibility as you somewhat alluded to. Do you think this is misguided on my part? Basically just so I can point at it in the future and say "a professional in the industry thought my book was worth printing with their name on it"

Not published yet sadly. I'm close to signing a contract with a publisher that would see it get released in the summer. I've contemplated self-publishing though so it can see the light of day sooner

Yeah, I think the least controversial takeaway is that you can afford to have a large protein shake immediately after workout. I personally don't think you need to. But this study and some others indicate that muscle responds best to protein consumption when it's in some post-exercise state (but it's unclear if that state is 30 minutes or 3 hours or longer etc). I'd say it's more important to focus on your overall daily consumption rather than the timing of it

Wow, I've only read the abstract so far but this basically goes against all conventional research on this topic AFAIK. I read probably ~50 papers on this topic (maximal protein dosing) and emailed a few researchers about it as I was writing a book. I believe the saturation point for a relative dose for a single meal is generally held to be ~0.25g per kg of bodyweight. Of course, there are some obvious observable issues with this belief given that there are people who practice intermittent fasting and have no issues building muscle, so it'll be really interesting to see if this study replicates

EDIT: After skimming the paper, I don't see anything immediately wrong with it. But there are some important nuances to note: the subjects were all fasted and given milk protein (casein in milk protein is known to take longer to absorb than pure whey protein which is a popular choice for these studies), and the measurements were made after an hour of exercise. This would skew the results towards more protein sensitivity than in normal settings where a person is pretty much always somewhat well-fed and not always eating after exercise. This is still encouraging because the results for post-exercise protein metabolism have still indicated a much lower limit than a 100g dose. Their report that oxidation rates didn't increase significantly is also notable since the belief has generally been that excess protein is oxidized and burned for energy instead of being incorporated into muscle. However, it would've been nice for them to include a 50g group as well to see if the dose-response relationship was really still linear between 25-50-100

Ultimately, this result seems encouraging for increasing post-exercise protein consumption for muscle gain, but we shouldn't discount the fact that the subjects were fasted before exercise. It would be interesting to see this study prolonged over the course of a day with further protein ingestions to see if the area-under-the-curve of muscle protein synthesis would eventually equalize in both groups, or if the larger immediately-post-exercise dose made a lasting difference. Existing research seems to not indicate such an "anabolic window". I might speculate that there is a daily limit for protein ingestion, but it doesn't matter if you hit that limit in one meal or five. That said, I have previously come across a paper that found medium-sized, spaced out doses to be more effective than infrequent large doses and overly frequent small doses, so there's still more to discover here

That's not what it means. This is a style of writing common in some academic fields where when they measure multiple qualities, they'll present them in the abstract next to each other in parentheses rather than repeat full sentences.

You should read "All else being equal, those highest on cognitive ability experience a 22% (53.2%) increase in the probability of realism (pessimism)" as as a 22% increase in realism and a 53.2% increase in pessimism

I haven't read the paper beyond this one section - but I plugged this question into GPT-4 and got a similar response. However, if I used military time (replacing noon with 12:00 as well), then GPT does get it right. Granted, it still hedges much more than any normal person would. But basically I wonder if it's struggling especially with the 12-hour clock concept

Ex Jane Street, Twitter, 6 YOE

Location: Los Angeles

Remote: Either way

Willing to relocate: Probably not

Technologies: python (django, pandas), scala, AWS, GCP, MySQL, PostgreSQL

Resume: http://blog.austinsteady.com/resume.pdf

Email: asteady23@gmail.com

I'm a backend engineer with about 5-6 years of experience. I've been programming since I was about 13. Recent side projects have focused on using GPT to automate fuzzy things like scraping Twitter and dealing with customer service bot menus. Over the last year or so, I also read a few hundred papers related to strength training and turned that research into a book on climbing training, which I'm currently shopping around to publishers

Embarrassing / dumb q, but how do you actually get to talking with them in a meaningful way? I can imagine with repeats, it's easier to build up rapport, but I don't think I've ever hit it off so strongly with someone at a coffee shop that we exchange info after one interaction (mine here aren't super social fwiw)

Thank you for articulating this. I remember similar problems and arguments arising after RNNs and CNNs became massively successful. People argued that training larger models would be infeasible for several reasons that all were made moot by Attention Is All You Need. Somebody seems to always figure out a new approach

I shared the code in a comment below (https://github.com/purpleladydragons/ai-sanity) but

- scraping is relatively dumb and straightforward. I use playwright to login and just scroll my timeline for the first 100 tweets. I run the thing every 3 hours right now, but definitely could tweak the number of tweets vs frequency

- I only care about AI tweets really on my timeline, so filtering to that is pretty straightforward just passing it to GPT

- I included the prompts in the link. Definitely far from perfect, but it works well enough. It does surprisingly suck sometimes, like I've noticed that it doesn't always pick up on tweets about LangchainAI despite AI being in the name etc

Sure thing - but just FYI I really only care about AI tweets in my timeline, so YMMV if you want it to summarize your timeline in general. I tried a generic "summarize my timeline" approach before and it didn't work well. Filtering to within a certain topic seems to really help

https://github.com/purpleladydragons/ai-sanity

I run GPT on a cronjob to summarize my twitter timeline for me. I'm sure it's not perfect, but it catches a lot

I’ve been fiddling with midjourney recently. There’s definitely a learning curve to it, but ironically gpt has been helpful for generating prompts. I expect the edge that artists have in this regard to erode further with time

If you are a manager, this will output your productivity ten fold on the upcoming years. Now you don’t need to hire senior devs and can just build the product of your dreams with very limited capital.

If you are a CTO, this will output your productivity ten fold on the upcoming years. Now you don’t need to hire managers and can just build the product of your dreams with very limited capital.

If you are a VC, this will output your productivity ten fold on the upcoming years. Now you don’t need to hire anyone and can just build the product of your dreams with very limited capital.

Agree it'll definitely be amazing for creatives and solo founders, but how many ideas are really out there to be had compared to the reduction in workforce?

https://twitter.com/paulg/status/1600119268858744832

FYI these are links to an interview with Ackman from March 2020 when covid was starting. He claimed "hell is coming" amid the initial 20 - 30% decline, right before the sharp bounce up. So essentially, maybe don't over-index on this guy's claims since he has his own position. (Thought it'd be helpful to add some context so these links aren't just dangling)

I'll offer some different advice from all the "quit" people (9 comments so far and all advise quitting).

My first job out of school was at a really top-notch place. I grew a lot early on but growth started to slow a bit already by 1 year. I became frustrated with things and wanted to leave. In retrospect, I still think that was probably fine / the right move for me at the time - but what I'd really like to have known is how to make better of the current situation. I lacked the emotional maturity and perspective to take things more into my own hands. Admittedly, I've gained most of my wisdom from job-hopping and seeing things at different places.

I would advise you to be prepared to leave, but have an honest conversation with your manager about your frustrations (and also what makes you happy there). If it's a good place, they'll reciprocate by helping you towards your goals. If not, then you tried, and you can feel good about leaving.

Also just in general, I find average tenure is a very strong indicator of things. Crowd wisdom is important. If most people leave after X time, there might be good reason for it. If your company's average tenure is much longer, then you should try to figure out why those people are happy to stay so long

Yes, there's a lot of hype around AI right now, and yeah sure it looks like some crypto grifters rotated into AI - but overall, I really don't get how people can look at all the progress and think there isn't something important here?

I see a spectrum of responses to GPT, but the really important thing is the rate of change. Even if GPT somehow doesn't impress you currently, why do you think things won't be improved in 6 months? What about 5-10 years from now? I can understand thinking that transformers are maxing out on their sigmoid curve of progress, but every time people complain about perceptrons, RNNs, CNNS, etc - another team invents something new and blows out all the benchmarks. These things won't get worse, so even if you assume a pessimistic rate of change, how different will things look soon?

The observation that these AIs are "mere text generators" is a testament to how insane they are IMO. The fact that they're just predicting the next token and are still this powerful is paradigm-shifting for how I viewed human intelligence. I don't think these things are sentient, but if they can replace me at work in 10 years or less, who cares?

GPT has been great for exploring new frameworks too. Copilot is amazing when I know what roughly what I'm doing, but GPT is like personalized framework / API documentation at times

Wanted to hop on and say this is amazing, thank you for sharing this! Also agree that it seems that it's really good at finding literally similar sounding songs, but not what I would expect a friend to recommend (this is both good and bad I guess). As someone else said, this is already way better than my spotify recs

Location: Los Angeles

Remote: Either way

Willing to relocate: Yes, coastal

Technologies: python (django, pandas), scala, AWS, MySQL, PostgreSQL

Resume: http://blog.austinsteady.com/resume.pdf

Email: asteady23@gmail.com

I'm a backend engineer with about 5-6 years of experience. I've been programming since I was about 13. I've worked across a spectrum of companies from seed stage to 10,000 people - including Jane Street and Twitter - which has given me a useful perspective on healthy and unhealthy orgs and what good growth looks like. My work experience has been in API development, ETL pipeline tooling, and devops. Personally, I spend a lot of time automating things in my own life and tinkering with ML

> modicum of status

It helps meaningfully with getting interviews at places for those of us who didn't go to top tier schools

As a late 20s, recently laid off, I'm wondering why you think it would be disastrous for your career? I'm not necessarily disagreeing, I've been in a very similar headspace before, but now that it's actually happened - without my choice - it feels a lot less bad than I expected.

Maybe I'm in denial, but I'm not super worried about getting another job right now. I'm enjoying the time off and reevaluating my priorities. There's even a work-related benefit: the extra time / lack of stress has allowed me to read up on tools etc that I haven't been using (although lately I've been tuning out of career stuff entirely).

I'm pretty sure the GPT model is huge and does not fit on any conventional GPU. Even if they open-sourced the weights, I don't think most people would be running it at home.

Also regarding the text limits, AFAIK, there's just an inherent limit in the architecture. Transformers are trained on finite-length sequences (I think their latest uses 4096 tokens). I have been trying to understand how ChatGPT seems to be able to manage context/understanding beyond this window length

Similar situation as you. I stopped keeping up around 2015ish.

I just read Francois Chollet's Deep Learning with Python and found it to be a fantastic high level overview of all the recent progress. There's some code, but not a lot. I mostly just appreciated it as a very straightforward plain-language treatment of RNNs, CNNs, and transformers.

Now I'm going through Stanford's CS224 lectures.

I'm sort of planning to read papers but as some other comments have pointed out, I'm less sure of the ROI on that since I'm not sure how feasible a future in AI is for me

Maybe start out not editing and just publishing your raw notes if you can. If people express interest, that might incentivize you to start editing. Basically prototype your writing.

Also, I find editing easiest with some time between writing and editing. Time away allows me to digest my own thoughts more and also makes my writing less precious to me. I sort of become a reader to my own posts and find it much more palatable to edit at that point