Future of programming with AI | Dario Amodei and Lex Fridman

[@LexClips]

Lex Fridman Podcast full episode: kzbin.info/www/bejne/q5jZeXaOeLSgo5Y Thank you for listening ❤ Check out our sponsors: lexfridman.com/sponsors/cv8229-sa See below for guest bio, links, and to give feedback, submit questions, contact Lex, etc. *GUEST BIO:* Dario Amodei is the CEO of Anthropic, the company that created Claude. Amanda Askell is an AI researcher working on Claude's character and personality. Chris Olah is an AI researcher working on mechanistic interpretability. *CONTACT LEX:* *Feedback* - give feedback to Lex: lexfridman.com/survey *AMA* - submit questions, videos or call-in: lexfridman.com/ama *Hiring* - join our team: lexfridman.com/hiring *Other* - other ways to get in touch: lexfridman.com/contact *EPISODE LINKS:* Claude: claude.ai Anthropic's X: x.com/AnthropicAI Anthropic's Website: anthropic.com Dario's X: x.com/DarioAmodei Dario's Website: darioamodei.com Machines of Loving Grace (Essay): darioamodei.com/machines-of-loving-grace Chris's X: x.com/ch402 Chris's Blog: colah.github.io Amanda's X: x.com/AmandaAskell Amanda's Website: askell.io *SPONSORS:* To support this podcast, check out our sponsors & get discounts: *Encord:* AI tooling for annotation & data management. Go to lexfridman.com/s/encord-cv8229-sa *Notion:* Note-taking and team collaboration. Go to lexfridman.com/s/notion-cv8229-sa *Shopify:* Sell stuff online. Go to lexfridman.com/s/shopify-cv8229-sa *BetterHelp:* Online therapy and counseling. Go to lexfridman.com/s/betterhelp-cv8229-sa *LMNT:* Zero-sugar electrolyte drink mix. Go to lexfridman.com/s/lmnt-cv8229-sa *PODCAST LINKS:* - Podcast Website: lexfridman.com/podcast - Apple Podcasts: apple.co/2lwqZIr - Spotify: spoti.fi/2nEwCF8 - RSS: lexfridman.com/feed/podcast/ - Podcast Playlist: kzbin.info/aero/PLrAXtmErZgOdP_8GztsuKi9nrraNbKKp4 - Clips Channel: kzbin.info *SOCIAL LINKS:* - X: x.com/lexfridman - Instagram: instagram.com/lexfridman - TikTok: tiktok.com/@lexfridman - LinkedIn: linkedin.com/in/lexfridman - Facebook: facebook.com/lexfridman - Patreon: patreon.com/lexfridman - Telegram: t.me/lexfridman - Reddit: reddit.com/r/lexfridman

[@extrememike]

Most LLM generated code needs to be revised, adapted or corrected for the greater task at hand.

[@Paul_Marek]

That’s most likely because the prompt was weak and didn’t have enough context. Accuracy increases dramatically when Sonnet especially gets good context.

[@vanderer_prashant]

How come you're still living in April 2024? Do you have a time machine?

[@mjcgreatness]

@@extrememike you can tell who the actual programmers are in the comment section. Ai is great for generating code for simple programs but it struggles mightily with more complex applications

[@vanderer_prashant]

@@mjcgreatness HAHAHAHA, you're in for quite a shock when you realize the actual reality 🤣🤣🤣

[@mjcgreatness]

@@vanderer_prashant when I asked what LLM you used for coding you couldn’t answer. You clearly aren’t a programmer. So zip it

[@jamiethomas4079]

Agriculture cant be disrupted anymore than it already has. We are down to single people and families running entire farms. 1 person is responsible for feeding 1000’s in some cases 10,000 people. We have squeeze the profits out of farmers so much they cant afford help. I am one of them. I need 2 employees here but I can’t afford it. AI may actually help us get more work done when used correctly. But then people will just demand more out of us. The general public has no clue where their food comes from and how its done to just a few people making enough food for the many.

[@brodiot326]

Thank you!!

[@Flibbinstein]

There are shortfalls in current Ai abilities in coding but you can't deny how powerful it is when a musician creates an app with Ai that he could never build on his own that turns around and makes 100mil dollars.

[@technophile_]

Here's my view on this (I'm just a normie, so please correct me if I'm wrong). I see LLMs (in the context of programming) similar to FSD on Teslas. FSD is already extremely capable, it's already 90% there as far as I know, but the last 10% is gonna be extraordinarily difficult. Covering all the edge cases that arise when driving a vehicle is a very difficult task and will take a long time. This is the same reason why we still have pilots flying planes even though most of the flying is taken care of, by the autopilot system. No matter how much we automate, we would always need human supervision (at least in the foreseeable future). Similarly, you might be able to generate code for a lot of common tasks in a very small amount of time, but, what matters is, when there is an edge case that AI is failing to help you, how long will it take for you to fix the issue. If you are an experienced programmer you'd be able to fix it much quicker than a person who just relies completely on AI and has very little knowledge when it comes to coding.

[@djpuplex]

It won't it can't. Until it does. AI is scaling all the naysayers are coping.

[@mattwyd2161]

Fr

[@54365100]

Well that's a way say that you've never done any decent piece of code in your life

[@j2csharp]

Yes, I think that devs will get back to being able to deliver more (as they used to be able to do). So much of programming is an incredible amount of abstraction and plumbing. Often, I'll find a problem with a story and have to go back to the customer and get clarification. But there may be times when I make an assumption on their behalf just to get something delivered that week. If I'm not bothered with plumbing code, then I'll have more time to handle the scope creep. How much of our jobs as devs are spent prioritizing quantity over quality? A lot, I bet.

[@bensaltigerald6304]

We are screwed when AI makes it to our fish

[@mirzakadic9174]

Yeah, trying using AI for something that there is little to none documentation online, like Adobe SDK plugin development. Good luck with that :)

[@MichaGero]

20 years programmer and using ai extensively. 100 %? These people live in lalaland. Fix your current ai to not forget code it wrote a minute ago, check current documentation, stop inventing parameters for existing apis, stop it making the same mistake over and over, stop letting it going down rabbit hole when there’s much easier way to do, and the bigger picture, ho boy, it has no idea, absolutely no idea of what it is trying to achieve. At the moment it’s good at code snippets, forget the rest. And the worse is 4.o has been getting worse in the past couple of months, to the point that I canceled my subscription. That’s the state of thing right now dear viewers. See you in a year and we’ll see if you’re still in your lalaland. EDIT: Right after writing this post I prompt GPT: "Why have you removed } else {" and it responded : "You are absolutely right; I inadvertently removed the else blocks". 🥳

[@dynamisticality998]

emergent properties from more compute ? grok 2 is pretty good

[@mattizzle81]

Yes it is both useful and infuriatingly bad at the same time. It is strange.

[@camronrubin8599]

You'll be impressed at the very least over the next 10 years . Your mind might even get blown .

[@its_ot]

@@mattizzle81 kind of like...... people

[@mattizzle81]

@@its_ot Nah. No more like people than a calculator. Much faster than a human definitely, but the types of mistakes it makes aren't human-like at all.

[@jamessullenriot]

Used AI to help me write some unit tests today. Spent over an hour prompting, re prompting, rephrasing, trying different approaches etc. Finally gave up and wrote the tests myself and it took me about 25 minutes including some twitter distraction. Its really great write now for boilerplate and extremely basic tasks, and what he is saying will be true at some point, but right now its just extreme hype from a CEO hyping yp a funding cycle

[@ryanmitchell2160]

What is the deal with the lights behind Lex? Are those lights the reason Dario kept looking to the side? Am I the only one noticing this?

[@mjcgreatness]

Ai generated code is trash unless it’s something basic that can be found on Google. Which is still a great use case, but LLMs have so many disadvantages when it comes to programming, we’ll always need a human in the loop approach

[@vanderer_prashant]

How come you're still living in April 2024? Do you have a time machine?

[@mjcgreatness]

@ I’ve used LLMs as coding assistants everyday for work since they came out, and they still suck in present day. What do you use?

[@jwolfe890]

His timeline does seem a little optimistic. Of course AI will eat us all in the end, but who knows how long we have.

[@johnbilodeau6296]

biased and optimistic opinion. I thought like that at the start of the year but it’s less and less the case. I am more and more disappointed with the help provided by the use of llm. At times it's great but it can be a waste of time and cause a certain amount of laziness. their development and improvement risks stagnating. a new approach will likely be necessary.

[@RawrxDev]

I feel like due to the novelty, people are isolating AI coding assistance conceptually, such that they are comparing programming with AI vs programming without AI (or any other assistance) which, while would have been fine 40 years ago, is much less impactful then people realize, when you understand that google has been apart of the workflow for decades now, and so AI's current impact is that of speeding up a google search. It also depends on what you are doing, it helped me a lot when learning react, and has helped out a lot less while learning graphics programming. I still find the quote I saw a while ago true, which is that "If AI makes you a 10x developer, you were either a .1x developer before, or you are working on a .1x project"

[@godblessCL]

I dont think so. It helps and will make programmers ... worst. The best think by themselves.

[@pandoraeeris7860]

Programmers are boned.

[@technoutopia4687]

This is why I switched to trucking. Coding is dead

[@aseeralfaisalsaad]

😂😂😂 I wonder if u have ever coded something useful or coded in general lmao

[@pequod4557]

@@aseeralfaisalsaad What does that have to do with his decision? Trucking is less stressful and more peaceful anyway.