orzig

Look, I basically agree with the goal and am trying myself but: Try watching the Little League World Series on YouTube twice weekly for 4 years and you will have 1% of an understanding of why this is hard.

I have not used it, but every review of the “AI software engineer“ DEVIN has pointed out that the marketing claim is absurd but the user interface is pretty cool.

That's fair, it might not be for you. In 'old school ML', for a binary classifier, there's the concept of Precision (% of Predicted Positive that's ACTUALLY Positive) and Recall (% of ACTUALLY Positive that's Predicted to be Positive).

It sounds like you want perfect Precision (no errors on specific Qs) and perfect Recall (comprehensive on general Qs). You're right that no model of any type has ever achieved that on any large real-world data, so if that's truly the threshold for useful in your use cases, they won't make sense.

I also had to build intuition for when it will be appropriate versus not. It's hard to describe but one very positive signal is certainly "will any hallucination be caught in <30s"? Even in ChatGPT Plus you can have it write its own unit tests and run them in the original prompt (even in the profile's Custom Instructions so you don't have to type it all the time).

So a mistake was using it for something where runtime performance on dozens of quirky data files was critical; that nearly set my CPU on fire. But str>str data cleanup, chain of simple API calls, or some a one-off data visualization? chef kiss

Exactly. Example: maybe 1% of the code I generate is bash. I used to try to memorize patterns, but of the top 20 I'd use each less than once per year. Now, instead of that 1% taking 5% of my time, it takes 2%. It's all "simple stuff", and I can verify it instantly.

I have ~10 similar use cases. So it hasn't revolutionized my life, but it's been well worth $20/mo ChatGPT Plus and $3/mo API calls.

Biology has probably had the most replication scandals of the hard sciences, but if you look back 10 years, there have been several tangible breakthroughs. CRISPR, mRNA vaccines, Cystic Fibrosis treatments, malaria vaccine, and that’s just off the top of my head.

So let’s not throw the baby out with a bathwater on Science.

I would be interested to see what other peoples comparable list would be for the social sciences – I don’t know enough to say that “absence of evidence is evidence of absence“

Adding a little bit of optimism: there are many decent people in all of these fields. They might not as often make the news (initially for extraordinary “achievements“ then for downfalls), but it is entirely possible to “make it in this world“ (supporting a family, taking the occasional vacation, retiring at age 65) while being honest.

Can we aspire to some baseline level of truth seeking where a phrase like “literally explicit” requires even the slightest bit of evidence?

In this case, it wouldn’t have even been hard to find that evidence, though you are assertion is oversimplified:

“ for the vast majority of the population, most of the pain of tax compliance could be eliminated by a few keystrokes at IRS headquarters. So why don’t we do it? Two reasons. One is lobbying by the tax preparation industry to discourage states and the feds from developing easier tax-paying systems, as California recently did. The second is lobbying by anti-tax conservatives. When the Golden State implemented its ReadyReturn system, it did so over the objections of Grover Norquist and his anti-tax pressure group Americans for Tax Reform, which fears that if taxes become less annoying voters might be less unhappy about paying them.” https://slate.com/business/2012/04/grover-norquist-and-h-r-b...

All of this took me 120 seconds with Perplexity AI. As we strive towards being a better nation, or at least a better community, I beg that future micro-polemicists invest that much time before hitting “post”

I loved the book The Billion Dollar Molecule (Reviewed by Nature here: https://www.nature.com/articles/nbt0594-521.pdf) about the founding of Vertex Pharma. Really drives home the amount of money, time and luck it takes to make a breakthrough - the author needed to write an entire second book to get to the part where they started making real revenue. The second book isn't as good, but if you like the first enough you might try it anyway.

Do you think the solution is better software, or is real blocker on the business/policy/cultural/incentives side? Inspired by that series I looked into building a better front end for clinical trials.gov but after seeing a few organizations try similar approaches without (yet) nailing it, I became convinced that I’d become one more classic naive-technologist story.

But have the existing approaches really missed a low hanging solution, even if it’s partial and maybe impossible to monetize?

Article is paywalled but the paper it’s based on is freely available here: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4675401

The author also blogs his latest Python projects (mostly Bayesian data analysis) on https://www.allendowney.com/blog/ , which I have really enjoyed.

Surely it’s a matter of degree through? I was at a company with a recently acquired-in product line and the amount of chaos on release day each month was notably higher. I would have described the original product’s development process as awkward but way less bad than the new one.

Having gone through the same cycle, I've done some amount of acceptance. HN is least bad, as long as I don't get sucked into writing comments (oops). The total amount of time I 'waste' is really not huge, so I try not to stress about getting it to 0. YMMV, but worth considering.

Neither are exactly what you asked but: 1) I did more than a year of freelancing work for a small company owned by a friend. It was necessarily stressful but I think the friendship was net positive. 2) I joined a fast growing team at a large company, reporting to a friend. There, the friendship was a huge plus because we can talk so frankly and get to the root of our disagreements. I think it’s extremely healthy for both of us.

If you’re in the United States, I promise you that any specialist you need to see will not have all your relevant data - I was once had a procedure held up because they could not get my weight imported from a different computer in the same hospital, because the electronic medical record was on different versions.

I will always upvote a good lessons learned post!

Love a published null result!

What prompt engineering do people use in real programming situations?

What's your annual salary divided by 2,000 (VERY roughly your hourly rate), and how many months of the service would it buy you as you try it out?

Totally understand if you're not employed, or barely making ends meet, but given the forum I'm going to take a shot in the dark here.

Can I open them with general software?

About 3 hours of dedicated feedback to the algorithm (including plenty of unfollowing, which is different from removing connection) has turned made it quite bearable. I don't _love_ it but OTOH I appreciate a place that I can be reminded of my weak connections without doom scrolling.

That's fine, I appreciate it! We're all conversing in public, and I'm sure it will be useful to someone.

And besides, this information makes me even more happy with an opinion I already have - isn't that what the internet is for?

I'm in the US but I'll say for sure that things like this are about half the reason I haven't hired any employees for my LLC. I wish I could, but I'd be de facto hiring an accountant as well!

My client had most of their money in Silicon Valley Bank. If there had been any haircut they would have (appropriately) put their employee payroll first. That was a scary time!

A Steelcase Leap V2, $150 (and 90 minute round trip pickup) using Facebook Marketplace. I probably got a little bit lucky, but it is definitely worth checking for things like this.

Also, a shout out to my friend who let me try his for 30 minutes since I couldn’t in a store.

As a senior IC formerly in a similar situation, I’ll affirm this is brilliantly written and even deeper than it looks.

I’ll add: Many of the Indian ICs I worked with were fine, but there was at least one manager who was an absolute disaster and the American executives had very little ability to fix the situation. Given that financial capital is here, executives will be here, and so some part of the org chart has to bridge across cultures, which is going to be a slight friction that compounds over the years of the project.

I strongly recommend building a little instinct in yourself for "what if HN search could help with this question" - for books especially: https://news.ycombinator.com/item?id=21900498

And someone took it upon themselves to build a book specific version across all discussions: https://hackernewsbooks.com/top-books-on-hacker-news

I check that fairly frequently!

The model needs four buckets:

20-30: Needs mentorship, can live closer to the office

30-40: Has young kids, moved to the burbs

40+ AND Management: Developed management approach pre-pandemic, pre-Slack

40+ but IC: Could go either way

This is still HUGELY simplified, and please do not take offense if you are e.g. 30-40 without wanting kids. But "young kids" is just a huge difference in terms of the morning schedule and flexibility needs, especially as most families near urban areas have both parents working.

Great quote, I bet it was very true in 1989 (34 years ago!)

I'm not going to try to prove the opposite, but after some number of DECADES it's worth checking on followup research. Here are some good places to start:

https://ux.stackexchange.com/questions/30682/are-there-any-r...

https://skeptics.stackexchange.com/questions/4355/is-using-a...