I just tried o3-mini

[@Adityasaurus]

Sam Altman when Sam Ctrlman walks in

[@Unreal-Paranova]

Sam Ctrlman when Sam Shiftman shifts in

[@saeruji]

Sam Shiftman when Sam Enterman enters

[@Exoepxoe]

@saeruji Sam Enterman when Sam Escapeman Escapes

[@Torudson]

Imagine when Sam Superman walks in

[@martinsmith2948]

Sam Ctrlman when Sam Endman ends him

[@caleb...]

Regardless of how accurate it is, LLMs being able to do "PHD level work" is a nightmare because even PHDs have trouble checking other PHDs work, let alone anyone without a PHD.

[@MyzaZoo555]

yeah, Sabine said the same thing

[@monad_tcp]

wait a moment , PhD is a grift ?

[@michaeldavid6832]

There's no wisdom in a crowd of coders -- or their code.

[@TheNewton]

​@@monad_tcp only too fools that can't recognize how language terminology and deep hyper specialized expertise work together even though it affects them everyday.

[@TheNewton]

PHD translation has always been the issue, as is over use of thesaurus in order to pad word counts and or protect IP by making a work inscrutable. Now papers can be made even more inscrutable faster with AI to counter trying to be understood faster with AI.

[@randxalthor]

Damn, talk about a fast upload. My favorite answer so far from o3 is when it told me that it was an ozone generation machine when I asked it what the difference is between o3-mini and o3-mini high.

[@young9534]

lol I googled o3 ozone generator and that is a real thing

[@Kwazzaaap]

Mine asked me for more context about what an "o1" and "o3-mini-high" were. Asking for context is better than being confidently wrong, but it doesn't feel good when you insert 1 of your 50 coins, and it asks you for one more coin.

[@young9534]

@ I wouldnt ask the models for information about themselves. They will hallucinate shit

[@user-sq1oi9qp8w]

o3 is whort form for ozone

[@mklabtech]

2:10 - try it. A stream where your wife without coding exp would try to make a working program using AI (probably Cursor is even better for this case). That would be an awesome content and experiment on the current stage of LLMs

[@Kane0123]

I wouldn’t want my wife to be on stream with the degenerates in this community…

[@cat-le1hf]

OpenAI would probably send their hitmen after them for exposing how worthless this is to non-experts

[@JohnDoe-ev8fk]

Fantastic idea. Have chat write the spec sheet so it's like a typical customer. Schitzo who has no idea what they want and changes their mind

[@grise123]

Yes man, do that

[@ezbody]

AI does a pretty good job when: 1. You don't know what you are doing 2. It's short Anything over like 1500 lines of code and it starts behaving like a drunk programmer, the longer is the code, the more inebriated it becomes, running in circles, aimlessly deleting and reinserting the same code, etc.

[@Definesleepalt]

its funny that most people who say that AI will replace programmers are people who don't write code

[@ThePurpleSnork]

It's not so different than hearing the phrases "Why does this cost so much?" "How long could this really take?" and "It seems like it should be easy!" from dumb people any time they're hiring skilled labor. All of us with skills of any kind have to hear that kind of stuff endlessly.

[@TheG7]

My friend who doesn’t code: Just ask it, then make it do it

[@Ivchiks]

Now let's wait for R2 😊

[@onatics]

gpt 5 is going to be out of the park

[@young9534]

I hate all the AI hype. But once you put the hype and sensationalized headlines aside, the tech is actually pretty cool. You just need to understand its limitations. I'm having fun playing with all the new models we've gotten over the least 9 months

[@Mylordkaz]

completely agree, I hateeee the hype 🤮 but I have fun playing around with the tech.

[@monad_tcp]

yeah, it is really cool, I spend a lot of time on HuggingFace

[@ShadmanSamin2004]

Finally a W take

[@Lemmy4555]

This statement is just perfect

[@noobicorn_gamer]

To be frank, same goes for image generation. The reason why we see so many shit images online (real or anime or whatever) is because people who have 0 knowledge in arts and design (yes, there are fundamental rules and color theories you have to acknowledge before breaking it to create something artistic) prompt whatever THEY feel like they wanna see and post it everywhere thinking it's 'artistic' or 'real' (fuck off with all those hyper-real, ultra-real, real is real, no need for stupid adjectives) which makes it seem that stable diffusion is crappy when in fact, when real artists use it, it can be a solid proof of concept tool, but the learning curve is so opposite from what creatives are used to, they don't touch it at all in most cases. All AI tools can be great for proof of concepts at the moment and people who are hyping over that it will replace people are super overblown. It's getting better for sure, but all these clickbaits by people who wish to get engagements are super not helpful, that's why people like primeagen and aiexplained, etc. are needed more than ever. Always appreciate your uploads man.

[@dingmana]

Love your hot takes on the new releases. Thanks Prime!

[@thingsiplay]

I don't know whats hot take here.

[@dingmana]

@@thingsiplay sorry guess I meant quick take?

[@HaydonRyan]

Honestly, that’s a super awesome idea for a video is to get your wife to build a simple website using cursor or o3 mini

[@BinaryCounter]

DeepSeek R1 is super impressive though. I'm currently building a compatibility layer to run some android games on embedded linux with very little overhead. I gave R1 a general description of what I'm trying to accomplish and where I'm at, and then told it to implement a portion of the Android NDK API (specifically ALooper) in C++. It looked up the documentation on that API and had a deep discussion with itself for around 200 seconds before writing out a header file and implementation, 1800 lines of code total. The whole thing worked flawlessly, first time, and passed the unit tests from the Android codebase, while being an entirely different implementation tailored towards Linux.

[@hck1bloodday]

I do't know if that's sarcasm or not

[@astraguardian]

i tried o3-mini that just got released in windsurf, but I switched back to claude sonnet 3.5 because it's still better and usable for coding

[@dude3278]

The year is 2038 people still prefer Sonnet 3.5 over o69

[@matej2714]

I guess the only real difference that should matter to most people using this is the price drops on these things. Claude eats like 10 cents per query (when it's not being bombarded and it actually works) so all the new models that came out recently that get a similar-ish output with some steering, and you have to steer all of them, for a fraction of the price is a nice change. One good thing I got out of this whole AI ordeal is being able to do OCR -> structured JSON for basically nothing by using models like Gemini or the free models on OpenRouter.

[@ErazerPT]

Yeah, that sums up my experience with them... The more you know about the domain, the more precise you can be on expressing it, thus the less "input noise" the LLM needs to deal with. And that comes with, yes, experience. From hitting some wall head on enough times and figuring out what you REALLY wanted to do. Take BFS/DFS for example. You don't even need to know how to code it. Just the fact you know what it is is a leg up on asking the LLM to do it.

[@crowsec-software]

I think ai's a good tool for going into uncharted territory. My guess is that people with solid fundamentals but less experience would get the most value from using these tools vs someone with a greater breadth of exp may find little value. Auto completes with context can be pretty awesome though. Having variables set. initialized, nice print statements etc in a single key stroke feels great.

[@icandreamstream]

No views in 5 seconds? Bro fell off

[@ThePrimeTimeagen]

I'm hanging up my keyboard

[@icandreamstream]

@@ThePrimeTimeagen😭

[@ProfessorRainman]

@@ThePrimeTimeagen How does that work with a split keyboard? Now I'm imagining one of those classic scenes of sneakers hanging off of a powerline, but instead, it's a Kinesis wired keyboard instead.

[@alst4817]

Found bloody and broken at the foot of a cliff? Bro fell off

[@plaidchuck]

The HTML and css it made was pretty damn impressive to me. Two prompts pumped out a modern looking site with mega menu and everything. Even faster than low code and Wordpress

[@tomfuller71]

Agree that o3 isn’t a particular milestone for code snippet generation - but disagree than the trend for planing AI’s isn’t clear. The objections that the code isn’t optimal, or potentially secure are valid. But It seems clear that a lot of money will be thrown at making systems of AI agents that can generate, then test and measure, then critique and plan code change to align to performant safe code. I don’t know if this will be realized, but it’s clearly a consideration if you are looking at software development as a career.

[@RogerDodger-fv6yr]

Yep, this is exactly what the trend is going to be. I've talked to big investors in the AI space, and they're already pouring money into startups that are trying to replace programmers with hundreds of little agents. The Windsurf IDE, which leverages agents, is much more impressive than anything like copilot.

[@Landgraf43]

I think they already have created some form of these agentic systems internally and are testing them right now to make them more robust before making them a product.

[@Brutalitops27]

The issue arises from new information coming into the space. Until there is a body of knowledge regarding a subject, an LLM would be extremely sub-par compared to someone who is engaging with the new information. Until you get something that can digest and work at the edge of the subject, they'll always be playing second fiddle to an expert/active practitioner.

[@tomfuller71]

@ Certainly with current LLM the lag between content capture and model release, combined with the limited frequency of novel training data hurts performance significantly. Copilot was terrible with generating code that works with the relatively new and still changing drizzle ORM package for example . But I’m not convinced these type of problems will persist. The frequency of model iteration seems to be decreasing. That combined with new methods of test time training that focus on ‘surprise’ as a factor, suggests that the benefit that humans have in a world of evolving knowledge is diminishing.

[@Dom-zy1qy]

I don't think we really need "PHD level reasoning" to be software engineers. Maybe if proof oriented programming becomes more popular, then it'd be more useful. What I want is a better base model without the CoT & test-time compute. I don't use copilot or even really use LLMs to write code, but I use it to ask occasional open-ended questions about my design or something I'm studying, and 4o just doesn't really give very insightful responses anymore. It just tells me what I already know. To get around that, I have to think for a little bit and formulate a more intelligent question, but in doing so, I usually end up figuring it out myself.

[@TheIshkrish]

I completely agree with your views on this point! I even encountered that with simple HTML and CSS outputs

[@sirrobinofloxley7156]

DS venture sounds exciting, good luck!

[@ryandury]

i love this take and the horse analogy

[@jacobgoldenart]

Well said. People are really discounting the main part of programming which not the hammer but knowing how to build the house.

[@sebacamposdev]

great pep talk

[@shahidullahmuffakir668]

so in the end : 'if you want to programme with ai, you have to be a programmer'

[@a__random__person]

idk... that's how i felt too. until i wrote a program / flow that has ai generate very detailed project, feature, task descriptions.. requiring everything to be expressed in Gherkin. Also created project outline, environment, tech specs, etc that I personally knew I wanted (didn't have to). All of this gets fed in to cursor compose agentic prompt, and automatically completes it without my interaction. I require everything to BDD, build, pass tests, run, and I'll even change a model from claude sonnet to a reasoning model if the ai detects that it's not fixing what it should. The result? It will basically create full platforms on it's own. I'm a backend dev, but also do web, android, flutter, etc (done lots of stuff in 20+ years).. shits getting crazy lol...

[@masheen_]

on that stringed html, gotta get that lit-html goin!

[@npc-drew]

this reflection highlight the relevance of TJ's points in that video he made about the hidden dangers of AI.

[@TheNewton]

Shibboleth, etc the fiction of techo-priests only able to operate artifacts through dogmatic litanies seems more likely to be based on reality everyday. On the other hand at lot of replacement hypotheticals also points at scenarios where soft-dev regulation isn't done for safety/quality reasons to turn the profession into a discipline but just to either keep people employed in order to climb the learning curve give without financial ruin, or make sure people are certified to access future mainframes either out of cost or restricted for knowledge control; or as an AGI joke ala Sigourney Weaver in galaxyquest.

[@TheBestgoku]

you are completely right, and I'm not arguing or trying to prove anything. But, before this technology, "noobs" would not even be able to take the first step let alone not able to reach where you reached after trial and error bcz of your coding experience of years. That first step taking ability is very amazing. I also understand that what you say is for coders and to calm their nerves around the fake hype and they should keep on learning coding.

[@denisquarte7177]

gave R1 and o3(mini-high) the same (small) problem, both overlooked the same issue and fixed in a second attempt. Nice, not impressed tough.

[@thematriarch-cyn]

When I tested out the model, it definitely seemed a bit smarter. Yesterday I had a bug in my code that took a few hours of talking with chatgpt to fix. Meanwhile, o3-mini was able to find the issue with only a little bit of hand holding. In fact, it made several assumptions (all of which were correct) that turned what would've been a 5-6 prompt long session into a 2 prompt long session. Definitely impressive, and better at debugging, but I'm not sure if its ability to code by default has gotten any better. It certainly seems to perform better in unfamiliar environments, which is mainly where I work(since i work on my own unique things, that generally havent ever been done before, or at least have only been done a few times). Impressive and useful, but ultimately not that much of an improvement.

[@luxeave]

If they think I'm slacking because most of my work is done by AI, they don't know the multi-pages of prompt I have to compose in order to get the exact result I need.

[@aubreyxengland]

i’m here for the mini-high

[@360Fov]

I'd like to locally host a code-oriented model for purposes of copilot style/autocomplete for rapid 'hello world'ing of unfamiliar lang. I've seen TabbyML suggested.... is it possible to run TabbyML on Windows with a Radeon? (bit of a longshot question I know).

[@westganton]

Once LLMs can replace good devs, they can most certainly replace your employer. Get those LLCs ready

[@UrFada]

So good to watch prime while cutting carrots

[@ArtursDerkintis]

Liking these shorter form videos - not every time I have the mental capacity to commit to 1hr video about subject I care to invest 5 mins

[@CosmicCells]

This is only the mini version... hyped for o3 and gpt 5 (which can apparently choos if and how much it wants to reason?!)

[@taureanwooley]

What was that weird stuff at the greyhound station in that one mountain city with the ai naming? I know some people came up with some stupid names that still need mentioning to save some people somewhere

[@AlanMitchellAustralia]

Even if you 'know how to program', do you know how to program everything? And in every language? Its those areas slightly lightly outside your circle of competence where AI can add value

[@morneauh]

Nice use case for Elixir Liveview

[@npc-drew]

hope I'm better understood why I told the YT algo to not show me any channels chilling o3 as AGI when it first came out...

[@cognitive-carpenter]

great end of video

[@jfndfiunskj5299]

Yeh, you're right. There's no substitute for knowing your shit.

[@moonasha]

I just used it to refactor a piece of code that was very ugly. I tried it a few months back with whatever model it had, and it gave me a god awful solution. The solution o3 gave me was elegant, and really cleaned it up well.

[@happyfarang]

Claude user her. It is very very very true that coding with AI is only valid if you know how to code and understand what it outputs. The fact alone that they have very limited memory about what have come before makes it a tool, not a replacement. If you can't vet the code it output you are lost. That said, sometimes, it just does what you ask for and spits out your classes and functions and i have no comments. It is also very good at finding stupid errors and help you debug much faster. When you can isolate the problem to a very confined box that is agnostic to the rest of your code base, that's where it works best IMO. But yes, when it comes to a larger picture, it fails. Memory is the biggest problem AI have at the moment.

[@thingsiplay]

Car comparison Its like with cars, the faster the car can run, the better your handling of it must be in order not to crash. That means, the faster and quicker and more an Ai model produces, the better you need to be to handle all of it without falling apart.

[@AccessCode101]

My test with o3 went really bad. The chat didn{t know what type of import to do in PHP when creating a plugin. Something so simple as to import the exact function.

[@e8p1q]

Did you ever figure out the hardware for your local llm rig?

[@user-qr4jf4tv2x]

Prime will be joining john connor

[@twezo]

Just said this yesterday prime, if you are good at coding these tools are great ngl but if you are more new, unfortunately you’re going to run into many headaches trying to understand the code the models give back. Just because it gives back something , like you said doesn’t mean it’s good code and everybody doesn’t necessarily know the difference.

[@tomb2840]

I was impressed. Although I feel it will be a while before significant trust can be placed in the code produced without your traditional team of developers. Just my opinion.

[@derpaboopderp1286]

This will never occur. The amount of ambiguity and interpretations of any sentence-> code is way too large. I might be huffing copium though

[@kraldada6557]

well there is the issue with alignment that noone solved yet. so until that's done it cannot work without control

[@skulver]

@@derpaboopderp1286 Indeed, we already have a tool that allows humans to express their intent in an simple, clear, unambiguous manner. It's called code. If you can't read through a program and point to stuff that you really did not need to specify you are almost by definition already at about as concise an expression of purpose as you could hope for.

[@user-pt1kj5uw3b]

If only we could, you know, read the code...

[@GarrettBlackmon]

I'm hype about it personally. I can program basically anything in any language if I had to with great effort and time. -I think that's inherently true of anyone really AI has made me more language agnostic which allows me to use the best tools for the job. AI says we should use go for this because assumedly a lot of other people have done something very similar in go? Great! I'm a web dev thats used node for everything my entire career but sure why not, the AI is now able to define a sensible project template for me, and as long as I know basic concepts like DRY, OOP and maybe time complexity (but lets be real, that has never mattered for anything I've ever built) and I chaperone the development we are cooking WAY faster that I could have done alone. Personally I think we are a long way away from "anyone with a good idea can make a good app" but when that day comes I'll be excited because it will push humanity as a whole forward. gatekeeping software development is understandable yet embarrassing. You had to put all that work in, and others don't I get it. But just take a step back and see we are in an exciting time that I'm so glad I got to witness.

[@asfdsdfasdf]

Pushing humanity ahead while hundreds of thousands of software developers lose their jobs so that money can be funneled into a select few. I’m in mechanical engineering and I’m confident as fuck it’s not getting replaced by AI in my life time. But whatever keep destroying yourselves

[@asfdsdfasdf]

and then gaslighting yourselves into thinking it’s a good thing. Now that’s what’s really embarrassing

[@cassell1253]

i use llm in my code but i do fully believe you should master the language before you do such things as if you cant understand yourself then a llm will misslead

[@acters124]

fastest stream and edit in the wild west dam

[@pp0l0]

o3-mini-high, which is supposed to be great at coding, still instantly shits itself with the quicksort function.

[@nicbeauieux]

Hey, I'm trying to improve my programming game, and I'm wondering where I can find a reliable source to learn good practices and how to write quality code that's maintainable. Any help would help! ** Free sources preferrably of course :)

[@Allplussomeminus]

Everyone can drive a car, but only a few can fix the car if something goes wrong. Learn to code yourself so you can spot errors and flaws in generated code.

[@chiragjoshi482]

hey prime.. great video. Saw the process live as well... i am a beginner programmer... you said in the beginning that it produced code, not necessarily good code.... what exactly does that mean? what about the code that it wrote was bad? if anyone could explain or give me any directions to see it'd be helpful.

[@cat-le1hf]

it probably did not actually execute or it did not accomplish the goal

[@Yugge]

Didn't see it live but I'm a senior engineer and I thought I'd chime in. Just looking at the code while he was scrolling by there were a bunch of "code smells" (look that up on wikipedia for more info). For example: 1. Lots of global variables. It just means it is using a lot of variables that can be accessed from anywhere in the program, which can be considered an anti pattern since there is very little structure to how a variable is affected and flows through the program. Which means that it is very easy to introduce bugs into the program if some function modifies it as part of their flow without explicitly showcasing a dependency on that specific variable. There are cases where this is legit though, logging packages may declare log globally so that you can add things to the log from anywhere. It does also mean that those values will never be freed from memory during the lifetime of the program, as they might be referenced at any time. 2. Inline HTML and JS. This is mostly just that you are including another markup language and inside of it a scripting language as a raw string. Which implicitly means that to the program, it just an array of data. You have no linting, error handling or other kinds of quality assurance tools into the html. If the html shows some kind of user data, it might be open to XSS attacks, as most routines for safeguarding against it requires the programmer to declare what part of the code that should be safeguarded against unsafe user input and what can just be rendered as HTML as it is controlled by the programmer. Probably a lot more, but I don't think I would notice them in the quick summary here, would have to catch the vod for that. In general though, it seems the code is functional, but carries a lot of bad design decisions that would lead to problems fairly quickly if the scope of the application would become bigger or if it had to be maintained for a longer time.

[@adriankovac1943]

O3s code did to many things at once making it not very adaptable to unforeseen edge cases and hard to debug and in a real world environment there are always edge cases.

[@williamcase426]

Oh snap it's out

[@CaptTerrific]

For full-stack work, I've been very impressed with o3-mini-high. However as soon as I get back to anything related to systems programming, o1 pro mode still leaves o3 dead in the water - FAR superior output from o1!

[@lule-ahmed]

Are u still paying 200 for it ?

[@puneetarora1714]

I just like to use cursor to search for code cause it indexes my code better than the inbuilt search, basically ask ai to search for code then go do your own thing and it uses less cpu

[@aship-shippingshipshipsshippin]

what if i can ask ai to help me out? i can say from the start that i have no idea about programming so please tell me the next steps i need to do and ai will do just that ;D im not saying ai is in position where we can just let ai do everything, no, far from it, but i think its starting to look like a lot of people will do stuff they cant right now thanks to ai ;d

[@PanopticMotion]

All AI can currently do-and will continue to do for a long time-is generate code snippets.

[@Chikowski101]

Are you friends with theo?

[@KingSvenDeluxe]

It would be awesome if AI could do my job, I'd be so much more productive and could build all of those fancy testing frameworks I have imagined. I just don't have the time to build everything and training juniors takes even more time. Problem is that AI is nowhere near ready, and it's just not going to be ready for a very long time. In my opinion, AI isn't even production ready yet.

[@paramdholakiya]

Is it me or he looks jacked af

[@ColourlessGreenIdeas567]

Fun Fact: We Are Not Coming. Do Not Hide, Beard Man

[@rogerballs3815]

You're using o3-mini. When o3 gets around, the leap should be great given code forces elo scores. But still, I think you're being slightly harsh or bias towards being a programmer. You may be feeling slightly attacked and wanting to defend your position. O3-mini in itself, let alone o3, is insane for programming. We couldn't remotely imagine this 5 years a go, it makes programming accessible for everyone, I know a few people who have coded programs for their business with zero prior experience using it. Would programming experience help you when coding more advanced things or when coming across bugs in its code? Sure, but that "issue" is dwindling with every iteration. We are at a level now where programming and creating advanced code is more accessible for a big jump in % of people. That's a great thing. It's hurting pure programmers who have dedicated big chunks of their life in learning how to program and its easy to feel annoyed by that, but its still one of the greatest advancements in human history and we are at a very high level already of usability.

[@Nico-qq7xl]

video idea let someone with no knowledge try to build something with AI

[@throwawaynyaa2616]

You don’t have to worry about AI taking your job if you are unemployed

[@mars009]

But I was told I could become a oncologist or a neuroscientist, they wouldn't lie would they?

[@Chris-se3nc]

I love these models because it makes my 19 years experience worth so much more. The new crop of devs is getting dumbed down 😂

[@raulavila-t5u]

Just wait for the mental loading, man. We're going all in!.

[@cohub77]

ur so old lol

[@jaysmitty-dev]

I really don’t get this perspective lol … yes AI is helping junior devs find solutions much faster but that doesn’t mean they aren’t learning from it as well. I had about 6 years of experience before AI and I’ve probably learned more in just this past year using AI to not only to find answers but explain in a way that resonates. Its potential for fast-track learning is vastly underestimated.

[@JavedAlam24]

Most people don't learn they just take shortcuts ​@@jaysmitty-dev

[@MrAmbrosse]

No, much less now, as what took you 19 years to learn can now be learnt in a lot less.

[@HolyChez]

garbage in, garbage out. You can't pretend that something is inventing beauty when it's just copying from a wide selection between beauty and slurry, randomly deciding between them all the time, and slowly being steered by human beings to what is desirable. As with all software, can be a very useful tool at best, can be a most malicious malware at worst.

[@archuser7607]

Waiting for r1 32b setup video

[@Mylordkaz]

I get frustrated when I use GPT... only good for translations. for the rest I am very happy with Claude.

[@jassingh6943]

hi primagen !

[@ashosaurus2140]

To anyone going all in on AI, genuinely, what is your moat? Knowing how to "structure sentences"? I feel like we all learnt that in high school. Where's the value if everyone can do it? maybe your moat and the value you bring comes from possessing difficult to acquire skills... like being a good swe? like having experience?

[@KristeHuynhnh]

Si j'avais autant de confiance en moi que le gars dans cette vidéo, je serais président à l'heure qu'il est💕

[@me-low-key]

que ce que tu yapping about???

[@gmonie619]

its better than hitting the pookie thats for sure

[@TheMightyWalk]

2 mins and 500 views bro is over

[@burnt1ce85]

you speak common sense sir!

[@npc-drew]

display the green screen to showcase he about to keep it real lol 😅👌

[@5t3w]

#CODEJOCKYS

[@ProfessorRainman]

I'm shocked I'm the fourth view. You should be trending already, Prime!

[@nrg4285]

Not saying this AI bs is a Porsche GT3RS but imagine putting a Karen, who's used to a Camry (Notepad++), drive a Porsche track car (30B parameter AI model)? She'd crash that thing in a heart beat

[@joshuaworman4022]

Wife thrown under the bus smh

[@HarshalGunjalOp]

Isn't this too quick.

[@timeiskey1]

r1 32b is not even r1. its the ollama distillation. it is horrendous at coding. its like beta copilot tier from 2021

[@quadmasterXLII]

I mean, they're all beta copilot tier from 2021

[@thomas6502]

Hoping these AI will do code reviews so I can offload the overhead of sifting through layers and layers of modern abstractions and be called in to "help" "guide" only on particularly egregious instantiations of "brilliance"... short of that: meh.

[@jeangiraldetienne8182]

This guy has a bias against Sam Altman😂

[@dev.sharif]

What he says in a nutshell: BRO just learn the damn programming, ai is not even close to taking our jobs. Even if it is taking our jobs it's 100% possible that we can use the ai to make a gap even further so then it's no longer taking our job. Also mentioning that ai is dumb and you need to walk with it so it produce functioning code REQUARES programming knowledge.

[@attribute-4677]

AI can’t even drive a car around without it getting stuck going in circles randomly, much less fully engineer software 🙄

[@nealgoogs]

This guy is coping lmao. Hes fine though. Hes been in the industry for so long that hell be fine. The rest of us just starting out are fucked. Had a homework for android mobile application for a simple math game. Literally one prompt and it worked.

[@giovannimaia9652]

Do you really think that your homework can be remotely compared to what it is to write software in real life? Good luck.

[@nealgoogs]

if you dont think the technology isnt going to improve, you also are coping.

[@plaidchuck]

He’s not even in the industry any more. Made his nest egg during the boom times and then retired early basically to be a KZbinr full time. I agree, it is disheartening that these things can basically do everything but enterprise level code. Makes small projects and homework pointless. You’re better off learning frameworks honestly

[@pedroleite99]

You should try to ask your wife to write some project using ai

[@chickenfondue]

I think the real risk that we should be worried about is not AI replacing programmers, it's AI replacing the program itself. Like the CEO of microsoft mentioned, when the end of SaaS comes and everything is replaced by AI agents, there will be much less need of programmers in the industry.

[@ryandury]

100%