Can You Trust OpenAI Press Releases?

Recorded live on twitch, GET IN
Article
asteriskmag.co...
By: Lawrence Chan
My Stream
/ theprimeagen
Best Way To Support Me
Become a backend engineer. Its my favorite site
boot.dev/?prom...
This is also the best way to support me is to support yourself becoming a better backend engineer.
MY MAIN YT CHANNEL: Has well edited engineering videos
/ theprimeagen
Discord
/ discord
Have something for me to read or react to?: / theprimeagen
Kinesis Advantage 360: bit.ly/Prime-K...
Get production ready SQLite with Turso: turso.tech/dee...

[@CaptTerrific]

If you can't trust a corporate PR team, who can you trust!?

[@edwardserfontein4126]

😂

[@MNbenMN]

maybe a Trust and Safety team?

[@carriagereturned3974]

you can't trust yourself))

[@black-snow]

world = crumbled

[@chrism3790]

I personally believe one day we'll be shocked at the degeneracy and irresponsibility that went on within Open AI. It'll be the FTX of AI.

[@cagnazzo82]

Comparing OpenAI to FTX at this point is a wildly off-base coping mechanism.

[@_Safety_Third_]

You might be shocked. I'll be rolling my eyes at the collective shock.

[@RillianGrant]

They have a product that works. You can't go too wrong with that

[@agastya2803]

I would be surprised if that’s what happened. Unlike FTX, OpenAI actually has a product that actually works

[@gameacc6079]

@@agastya2803 study amazon auto checkout

[@ika_666]

i have near human coding capabilities

[@sjoerdvogel3352]

I have near human social skills

[@edwardserfontein4126]

😂

[@AlgorithmAlloy]

I have near human making-my-girlfriend-not-mad skills

[@userDude4]

@@AlgorithmAlloy god its so hard sometimes xD

[@sam6000]

Best comment

[@Sunrostern]

Your mustache is near human,

[@_Safety_Third_]

My underwear are near human

[@hungrymusicwolf]

Beyond human I daresay.

[@afuzzybearsyoutubechannel2812]

some say tru AGI exists in primes tash

[@tttm99]

Near human - as in definitely near a human.

[@mikebierle1979]

Press releases are inherently marketing material, and US law gives wide latitude for marketing puffery. These are not scientific papers and the engineering department was likely absent from the team that wrote the press release. Remember, extraordinary claims require extraordinary evidence.

[@monad_tcp]

everything coming from the press is marketing, there's not journalism ever, in fact I don't even believe journalism exist, everyone has their own motivations

[@SimGunther]

Why wouldn't they fully commit and just write all they press releases about AI in their own AI if they're claiming it's soDangerous?

[@ripplecutter233]

Maybe they already do 😂

[@joni062443]

Omission of data/findings is fraud if they pertain to the context of the claims made. I would therefore conclude that omitting the negative results makes the overall claim fraudulent in this case between Claude and GPT comparisons. Of course I am not surprised. This is key to the scientific method's efficacy and reliability via separating data from bias and why why independent peer review is so critical to establish the most accurate current truth possible. The moment AI started to move from pure sci/research and stepped into corporate and monetization, this was inevitable.

[@TomisaMaker]

Omitting data = fraud. Negative results matter. Peer review is key. AI + money = bullshit. AI + people = metalic monkey + regular monkey = monkey strong together

[@letopizdetz]

Sam Altman on Lex podcast... gave some serious villain origin story vibes.

[@helloimnestor]

gonna call it jippity now!

[@sjoerdvogel3352]

me2

[@seancooper5007]

I can train AI models that are above Human-level ability. Humans in comas.

[@ender5023]

Me too! We should compare research notes my fellow colleague

[@the-answer-is-42]

I can make one that beats humans at math. Infant humans, that is, who hasn't learned to count yet.

[@monad_tcp]

only humans in comas ? what about humans in colons ? and semi-colons ?

[@_evillevi]

let's just let ai winter come already I'm so tired of llm hype

[@and1play5]

No, you're going to keep jerking off to Primeagen grifting off AI content while billions and billions continue to be spent in building out more data centers. Or maybe you're not going to be able to survive on this planet. Who knows?

[@young9534]

New models are supposed to drop from Anthropic, Google, xAI, and OpenAI in the next few months. And OpenAI is getting a lot more funding. So we are getting an AI Winter, just not the kind of AI Winter you're hoping for

[@user-jg9pn9wh3g]

Given the apparent risk for this field I instinctually also want an AI winter, but when I think about the huge prosperity that it would ultimately bring if such winter never happens it makes me feel guilty.

[@and1play5]

@@user-jg9pn9wh3g You can wish and pray for AI winter all you want. Jensen Huang doesn't care. We're gonna solve mad diseases. Nobody gives a f about your negativity. Remember that. But keep grifting.

[@vantage8294]

​@@young9534 what does that mean can you explain ?

[@M43782]

Chimpanzee is in 98.8% genetically identical to human. Banana is more than 60% identical to humans.

[@55sombreroman]

All the shit you have to do to make AI useful seems like more work than just doing the task yourself from the start.

[@midiminion6580]

Just yesterday I was trying to use GPT to write a couple of JS functions. Ended up wasting a heck of a lot of time, scratching all of the GPT generated code, and build mine in 30 minutes. It's a good improvement over Googling stuff and Stackoverflow, but that's about it.

[@iMagUdspEllr]

Yeah, like inventing typewriters, calculators and computers. But, the productivity increase is worthwhile. As long as it takes less time to check and review code than it does to write it yourself, it's worth it. It is worth it. I ask ChatGPT to write something or debug something and then I check it. It takes objectively far less time than it would to write it from scratch especially if you count all the documentation and planning I dinxt have to do. Of course if what was generated didn't fit the requirements, then I have to ask it to change part of the code. This often takes way less time than writing it manually. Don't worry, we'll have jobs because AI doesn't conform to business requirements, generates faulty code, doesn't create good architecture, and doesn't pick the right solution every time. But, let's not pretend you type at 300 WPM and can plan everything in .25 seconds.

[@calmhorizons]

@@iMagUdspEllr Interesting and I somewhat agree. But important to bear in mind that all of these technologies also create inefficiency. For example, computers and the internet enabled Emails. This was intended to be a huge efficiency gain over letters. In practice many now question whether Email is actually an efficiency drag, because lowering the barrier to sending a message causes people to send them pointlessly, spam them, CC every person in the company etc. etc. I get hundreds of emails a day that I don't need, but I need to review them just in case there is something important hidden in the pile. A complete waste of my time. Similarly, if AI enables people to write lots of shonky code and standards drop as a result, we may actually find ourselves worse off than now. Badly written code (IMHO) is a bigger drag on productivity than no code at all, because you have to maintain it, improve it, rewrite it, and so on - and any errors it creates downstream could be disastrous e.g. the Therac-25 bug that killed people with radiation.

[@SidTheITGuy]

AI tech never deserved any attention, and now Microsoft is putting $19 billion in starting up a NUCLEAR REACTOR to power the tech. It's mind-blowing.

[@abhinavrobinson2310]

Add on top of it the layoffs done to justify it.

[@and1play5]

Nobody cares what you think. Y'all don't even work anywhere. Just keep watching his videos and getting him paid. Get Primeagen paid. Because AI's gonna keep advancing. You don't see the money pipeline?

[@thekrustychub5038]

Ight relax dude, you can be bearish on large scale use cases but saying it deserved no attention is just a ridiculous statement.

[@worldspam5682]

Huh? Since when nuclear power plant costs 19 bil? Is it some bs "new type" of reactor, that is still up to be discovered?

[@young9534]

I'm all for nuclear energy

[@timmysmith9991]

OpenAI 49% owned by Microsoft? What could possibly go wrong?

[@GackFinder]

Indeed. Don't underestimate MSFTs ability to F things up royally.

[@rnts08]

At least it can't make it worse than they're currently making their wintento Spyware.

[@jerrygreenest]

Not an Microsoft fan here, but that's not the reason. OpenAI were frauds before Microsoft. In fact, being sold to Microsoft seems like fraud itself, they tricked Microsoft into buying this shit

[@user-pt1kj5uw3b]

I also think the prominence of pre-peer-review papers on Arxiv has led to a lot of sloppy research and likely fabricated or misleading results.

[@romangeneral23]

Can You Trust OpenAI Press Releases? Answer: No Reason: How else are they going to raise more money to burn for investors

[@connor_4132]

Altman just trying to get more money for his unprofitable dumpster fire

[@young9534]

OpenAI is getting another 6.5 billion in funding. They actually had to turn down some investors because they reached their funding goal already

[@Jesus228FBI]

Then we will get an economic crisis and the government will bail OpenAI out

[@Jaythre]

@@young9534cope

[@msc8382]

​@@young9534 6,5 billion? That's going to be a lot of money wasted for sure! The problem with funding is that you still need a profit in order to have freedom. With such a funding, they'll never profit again. All I'm hearing is a company that is going to be bought out by the government once its collapses. Riddle me this: if the company was so successfull with profits, why would they need additional funding? Addem, why 6.5 billion? Why not a few million? Something doesn't add up. It starts to smell like a money laundering scheme. I'm an avid chatGPT user, and I have access to it due to my work. Even version 4o is barely better than an intelligent lookup tool. But every day I have to adjust my prompts because it simply has no intuitive hair in its bones. No amount of virtual brain will solve that problem. 6,5 billion to solve a problem they haven't even started to address? Within 'a few thousands of days'? No sir, this isn't possible. That's like trying to engineer the modern education system... while books haven't even been written.

[@georgeokello8620]

@@young9534that’s corporate speak for we are hemorrhaging capital resources because we can’t create profit and we are going to other investors as the current pool of investors are shrinking as they don’t have capital aka funding to recklessly throw at an idea that isn’t giving them their return on investment

[@magfal]

My experience is that Mixtral 8x7B Instruct is within statistical error margins of GPT4/4O and O1 preview for code generation.

[@bencor4193]

That bioweapon tests at the middle hit quite a bit differently then the rest. Using AI to create new bioweapons is scary thought.

[@astroviking3414]

I'm in week 3 of a 3 year programming course . Glad to hear your opinions man !

[@jeremybobbin]

If most customers have an M2 chip, I don’t see anything wrong with “M4 is 50% faster… compared to M2”

[@mwwhited]

So wait... companies lie in press releases... you have to be kidding me.

[@PedroNacht]

Is the CoT vs non-CoT thing a problem with the individual companies' behavior, or is it just an indictment of the benchmarks? It does seem pretty insane that the benchmarks don't require a specific, reproducible prompt.

[@MiguelOliveira-wz4ju]

I trust the mustache man

[@VanjeAv]

The mustache has IQ bigger than the AI

[@jakubbartczuk3956]

This is so dope. I'm working with AI (I was doing NLP before it was cool) and I'm fed up with the empty hype. This article is on point, it focuses on ACTUAL problems with AI and how not to fall into marketing bullshit trap.

[@davidkosata3051]

One thing I don't see mentioned often enough is how easy the benchmarks are. MATH is just problems that a highschooler can do in 30s with >90% accuracy, and I think most other generalist benchmarks would be similar, while something very specific (some models are being compared on symbolic integration, like wth) is harder for a human with no training, with the same training dataset it would still be probably much better to ask a random teenager (with minimal wage).

[@my_online_logs]

the model is tested on phd level math and physics. just keep coping to say that a highschool can solve php level math in 30s. im very sure you cant solve highscool level physics in 30s

[@Testvvjnb-ci3zl]

​@@my_online_logs I agree with your take but... can it actually solve 'new' problems. realistically most of them are in the training data in some form already and solving solved problems won't bring us much further (but it may help in the discovery)

[@davidkosata3051]

@@my_online_logs idk about you but I went and solved some of the benchmarks manually (mainly MATH though), and I agree that it was a bit of a reckless generalization saying that all others would be that easy. But those that I tried were really that easy (I am a high schooler and I had like 96% doing it usually in the middle of the night).

[@orderandchaos_at_work]

Hell no

[@zyme5998]

funny how noone knows what copilot uses, it really depends on a lot of factors, for example on some of the Android apps it has an option to switch between 3.5 and 4, while in Windows 11 it use to say that if your not using Copilot Pro you would be be using gpt-4 except if utilizaation is high then you might be using 3.5. The last time I looked it said it was using GPT-4-mini, and I'm not sure if that's across the board, I rather assume that it again depends on resource utilization in their server farms..

[@fluffytux1732]

I can only explain Pime's worry about AI writing research code by a lack of exposure to code researchers write without help. Both are equally scary to me.

[@uuu12343]

When your CEO is called "Sam Altman", you cant trust the company in general

[@amesasw]

Been using o1 mini for dumb work questions. Still not sure why a simple question requires a formatted book like response. And it's code still has silly mistakes in it.

[@Derick99]

What about the super chip nvidia is building that they need a nuclear power plant to power it? Won't that increase ai to a crazy level just based off scaling

[@LiveType]

Yes. B200 has about 2.6x performance and about 4x if using a more sparse fp4 precision. Issue at the moment is supply. Apparently nvidia encountered some bugs with both the chip itself and packaging issues which delayed shipments by several months. It should be shipping to initial buyers at the very end of the year/early next year. 8 B200 gpus to fit into a server will apparently cost 240k which seems a bit cheap imo. I would have expected closer to 300k. Power draw per gpu: 1kW Power is creeping up and up and up. No wonder micrsocoft wants nuclear power plants to power those data centers. A single rack will eat up like 100kW!

[@ElmerGLue]

You also get diminishing returns at some point since everything in between must also meet demand and that’s where R&D gets stuck much of the time. They keep on saying we need more power but it’s coming at the cost of actual humans being able to use that energy. They are talking about nuclear energy now but before they just used up the fresh water in the desert to meet demand and cool off the heat. Now they want to exponentially expand this and we still have yet to see if all of this will truly lead to new knowledge or just regurgitation and wasted energy. Do we need a nuclear power plant to have a fancy search?

[@mathewszulman7659]

The issue is that drastically improving compute only gets accuracy up by smaller and smaller increments and benchmarks don't give you a whole picture as these llms are now being geared to getting scores on them rather than actual utility and function. When it comes to deconstructing human brains we also think we can just break down the 'neural network' but forget that this is only half the story, our brains are ultimately something that interacts In the biological realm, making a digital copy essentially takes out a huge portion of what makes it work the way it does.

[@Derick99]

Yes but meta and grok can pose some competition especially with all of the top people at open ai leaving the company? They are probably all being recruited and helping ilya. Yes it goes up smaller and more expensive everytime, but if you could double the capability that alone would be incredible, that plus self training and teaching ai is probably already here. Remember how long ago qstar leaked? Imaging what top private companies actually have cooking up. This plus data refinement and advancement in algorithms + robotics advancing, I would say gpt is probably already smarter them most humans not all but alot. There's litterally only like Sam left from the original crew

[@ElmerGLue]

@@Derick99 it isn’t smart though, it is a knowledge store and it is another form of lossy compression.

[@kietphamquang9357]

never thought "More human than human" slowly becomes a norm now

[@magfal]

23:23 this might be why I hear some people praise LLM coding abilities and i can never get the models to provide a net benefit above 5% regardless of which method I use.

[@thisbridgehascables]

AI in these models I about will reach general intelligence. Think how quickly the human brain deconstructs , analyzes , processes.. and then ask yourself.. do you believe these AI models really have the potential for anything but guessing? Also training AI on garbage data will never make it an expert at anything.. AI can’t do reasoning.. it can’t actually comprehend information and properly make creative solutions that would be ground breaking.. I think the method of teaching AI is wrong.. it should be modeled more on how you teach a human child.

[@rnts08]

Spot on, they'll just get better at predicting output and completing output quicker, there is a real issue now with lack of quality training data. Everything else is just fine tuning now. Increase context, more tokens and uploading private code and documentation.

[@my_online_logs]

@@rnts08 so you never know about the different architecture of llm. predicting the next token is gpt model, not all lmm, its the decoder only model, but its the gpt 2 as its the only version that is open sourced. we never know how the architecture of gpt 3, gpt 3.5, gpt 4o, and the new gpt o1. but given the difference intellegent of gpt 4o and gpt 2 is very huge, so they are very different, and the new gpt o1 is more intellegent than gpt 4o, that just mean the new gpt o1 is much more intellegent than gpt 2. where the encoder architecture try to understand tokens. and you think they dont have dedicated team to filter data aka just blindly feeding data to the model is a noob take. no way suct a giant doesnt filter their data kid. lol, it can create all things you said. you are just denial that you think it should 100% similar to human approach, like saying a plane should 100% similar to bird

[@Karn0010]

"Near-human", I can believe it given how politicians act the ai isn't far from that.

[@highvoltage3000]

You don't have to watch the video. Answer: No.

[@PassifloraCerulea]

Betteridge's Law

[@kdborg]

When I hear "near human", I translate it to "which human?" There are a wide range of humans out there and many aren't that smart.

[@chrish8941]

Love the vids. Boy oh boy I would be so happy if you could clean the harshness and rattle out of your audio

[@rikudouensof]

You can create a model with same class data and have high percentages

[@miindaugas]

"Near human abilities" its just funny..

[@young9534]

Not that we should buy into Silicon Valley hype too much, but I'm using the new GPT voice mode and o1-mini. Both are a big upgrade over what we had just one month ago

[@chrisjsewell]

Could you expand a little, are they helping you more on certain tasks or allowing you to do new things etc?

[@and1play5]

@@chrisjsewell Don't ask them how to use it. You all hate AI, so don't use it. Just type with your hands. Use a keyboard. Use a typewriter. AI is not for you. Remember, it's a grift.

[@chrisjsewell]

@@and1play5 go play in the sand and let the big boys talk

[@young9534]

​@@chrisjsewell Try GPT o1 mini and o1 preview and see how you like it. Its way better at debugging and writing boilerplate code than any other AI model I've used. I had it set up php unit with some basic unit and functional testing and it did surprisingly well. Ran first try with no errors, which is more than I can saw about other models. Also pretty good at math now

[@chrisjsewell]

@@young9534 ok cheers well yeh ill wait to see if/when it gets added to copilot; that has its uses, for autocomplete and reminding me off commands, quicker than a google search, but I’ve never really found it able to write anything too complex

[@alwayslearningtech]

I ran an experiment on an test to see how accurate Google Gemini was on a cyber security test. It was multiple choice and got 81% correct. The reasoning it gave for the incorrect answers was interesting and understandable,

[@auxwarzone6335]

It’s definitely better at some things moreso than others. I gave GPT 4-o a grad level compilers exam and it got 11%.

[@Iggysdust]

@8:49 that is CRAZY.

[@amesasw]

People understand that tests are not a good measure of human intelligence, right? It's just the best we can do in a brief evaluation to compare people.

[@fennecbesixdouze1794]

I don't know if this article mentions it, but tiny tweaks like shuffling the order of the multiple choice answers in the MMLU benchmark can completely reverse the LLM rankings chart. This whole industry is so massively fraudulent. Which is not at all surprising: OpenAI rapidly turned LLM research from an academic discipline into a big-bucks industry, essentially setting back AI research not just decades but potentially wiping out any hope for actual genuine reliable research progress as we know it. AI might have been big and might have had massive, life-improving effects, but now all the air for actual research has been sucked out and replaced by results-hording, lying, unintersting navel-gazing tweaks, and marketing posturing.

[@joseoncrack]

It's just a giant trojan horse.

[@kinuthiamatata6040]

So true....too much hype. Not biased

[@YaofuZhou]

For the discussion at 12:00, you can do even better by asking it not to do any guesswork, and instead add any debugging printours it needs, and show it the printouts.

[@H-xt8io]

Why do you use chrome? If you want chromium, why not at least use brave?

[@TheGuillotineKing]

If AI can solve customer service that's all we need because just that would make a huge impact

[@AlmondFlour]

now its Nearer-Human!

[@alexandermoody1946]

It would be very unwise to wish the warhammer 40000 universe into existence. The prospect of loss of all meaningful joyfully creation would be absolute.

[@gnuemacs1166]

Lobster is near human until u eat it

[@henry-js]

Can you trust Primeagen to not burst your eardrums with his mic volume?

[@evasivezim]

The BS remover hit hard on this one 🤣

[@Chris-on5bt]

On the fraud question: I understand orgs and corporations putting the best foot forward. Fair enough. I think its problematic in this case because of the comparative nature against another product. It gives the impression that this is some objective fair duke out between two parties, when really someone's thumb is on the scale. I think tech companies need to post all data about these new models, and then the journalist should *gasp* do their research and have unbiased reporting. When a corporation tries to get away with this crap they should be roasted across the board. But then again that sounds too much like a well functioning society where we have normal honest people operating things. My opinion, this fake economy will have to all burn down before any real progress is made on this kind of thing.

[@procrvstinvtion8479]

OpenAI's CTO just left and Sam Altman will be given equity as OpenAI will be restructured as for profit lol

[@BooLightning]

I’ve tried to tell people, but they still use llm’s as if they are 100% accurate

[@gerdokurt]

If Ford was a Tech company : Our new model is 3 Times as fast and efficient than previous generations !!* * Compared with model T

[@damiengilbert7055]

One way or another I want my Machine God.

[@tomorrow6]

First issue: can you trust any press release ?

[@seese9456]

you can't trust advertisements

[@BW-zq2tu]

Not allowed to train on benchmarks, if you could, it would be 100%.

[@gnagyusa]

AI presents facts. A lot of humans prefer their biases and fantasies over facts, so they down-vote it. AI is "upgraded" to not use facts, to get a higher score from humans.

[@p1rkarlsson]

"And palindrome!" Nerd

[@peterromfeld4091]

how can you expect correctness from an model that cant even do math?

[@Bodom1978]

I stopped using AI for the time being, why pay and waste time improving their models. Once they're ready to do what they promise I'll consider them again. For now it's just news to follow.

[@jichaelmorgan3796]

Hypotenuse, hypotension

[@_Safety_Third_]

Tangent, tangerine

[@albinekb]

No you can't trust them at all, they have "Open" in their name, but release no open weights.

[@MangoTheRetriever]

Near human mustache.

[@random_bit]

corporations lie, thats no secret

[@VelcorHF]

My favoritist angry people are data scientists type folks upset at AI hype. If I had 86% accuracy, I hope to god I’m not doing anything important.

[@christianrusso6100]

Ai has solved nothing. Ai has only caused problems in the economy and I think the only solution is for investors to give up on Ai because it is a lost cause. They already lost millions of dollars and they will only loose even more if they don’t pivot to a better strategy. Every dollar invested into Ai is wasted.

[@Testvvjnb-ci3zl]

what that is absolutely not true. in medicine for example it is starting to get useful. and we still have LLMs which are usable and helpful coding tools

[@IAmPattycakes]

I think losing millions is underselling it quite a bit. Openai alone is projected to lose 5 billion dollars this year.

[@leom8605]

@@Testvvjnb-ci3zl Only people who don't know how to code find them useful, most of qualified/certified Software Engineers find them an utter garbage.

[@zx-sy1qh]

Donald "They are eating the pets" Trump is human 🙂

[@xCheddarB0b42x]

Draw seven red lines.

[@tonysolar284]

Web 4 AI, here we come.

[@dgo4490]

I only trust an AI PR if written by AI.

[@solidsteal]

Dude, times have changed - you have a more sensible marketing strategy? Sell me this pen...

[@LengCPP]

I like reading with you

[@tablettablete186]

No... roll credits

[@Kane0123]

Certainly!

[@LookingAway359]

OpenAI CTO just resigned

[@tuhin1264]

I know how easy it is to lie with statistics 😂

[@bedro_0]

HOW MANY SECONDS AGO?

[@ika_666]

69

[@a.tevetoglu3366]

No.

[@KillianTwew]

9:28 or reporting things you fixed last time you renovated 10 years ago without mentioning all the things that have broken since lol

[@artscollab]

There’s a lot of fomo in these comments.

[@natyyyyyyyyy]

no

[@Twitch_Blade]

Yeah no ai gonna take all the jobs, definitely don't continue to learn to code!? (I for sure don't want more competition :D) JK learn it if you have a passion for it.

[@vncstudio]

GPT 4o is pretty good and better than 3.5 and 4! Very low hallucination. Prompts have to be specific and the problem broken down in stages for best results.

[@Trollhunter181]

Love from india ❤

[@chadjaax]

I can trust it at least more than I trust any Prime take or opinion.

[@and1play5]

Yeah, big time. Primes just making most of his money now talking shit on AI. He doesn't even make normal content anymore. The streets must be getting dry for Dev Twitter and Dev KZbin.

[@chadjaax]

@@and1play5 Rust and criticizing AI and JS is his whole personality now.

[@bnlbnlbnl]

Why do you guys even watch his content if you don't enjoy his content or respect his opinions lmfao

[@chadjaax]

@@bnlbnlbnl We enjoy the content, don't respect the Opinion.

[@WoodyJohnson-r3v]

Companies have always and will always talk up their products. However just like the iPhone as much as I dislike the product and as much as the company talks it up somewhere in the middle the iPhone is a powerful device that does amazing things. These models do amazing things already and LITERALLY NO ONE HAS SAID THEY ARE PERFECT. I really have a hard time believing that an intelligent person can't see that they will be doing most of the coding and at some point maybe a long time from now they will do all the coding. Do you really make that much denying reality?

[@WoodyJohnson-r3v]

@@TomisaMaker Yes thats my whole point. Just like the CEO's of Apple pretend the iPhone is perfect. The thing is that ppl make more youtube money pretending that AI is trash. Even right now we are creating engagement for PrimeTime. But in reality if dude is half the developer he claims to be then he knows as well as I do if you follow the improvements that has taken place since GPT 3 was first released just a few years ago it is VERY VERY CLEAR that AI is going to replace most if not all devs at some point. do I think that it will happen tomorrow no. do I think we will all be surprised how fast it happens probably. its likely to happen faster than even I think it will. Go use the o1 model and tell me its not shockingly good compared to GPT-4 or 4o. I have never done any coding except printing "hello, world" in python. I was able to create a whole web app for my daughter to use as an agenda book and serve it over my LAN in 20 min with 2 prompts it wrote every line of HTML, CSS, and JavaScript and the only reason I needed a second prompt is because I forgot to tell it that I wanted to add placeholders in the code so that I could just drop background pictures.

[@WoodyJohnson-r3v]

@@TomisaMaker Also, in the very article he is talking about in this video both the article and PrimeTime say AI can't replace devs and whine about devs loosing their jobs to AI. sure its the low hanging fruit for now but 2 years ago no one thought devs had anything to worry about and 3 years ago AI could never replace artists and writers and now everyone is freaking out about the insane quality of art and creative writing ability's of the model. sooooo, stay tuned.

[@WoodyJohnson-r3v]

@@TomisaMaker The UI? That's your argument the F*n ui? For one OpenAI has already implemented most of the lipstick you mentioned through custom instructions the GPT store and memory controls (the ability to delete the memory's it keeps). It is the Modular Tools so that is nonsense.you give it data and tell it to make a spreadsheet or translate. (the new voice mode can translate conversations in real time) you can upload, create and download documents in the chat for most models at this point so I don't know what more document handling you want. same thing with the mind maps you just prompt it to create a mind map and it will do it. really the only thing you mentioned that is not already available is the local data storage. If you want a different UI tell it to build you a webapp using its API and Bobs your uncle.

[@maciejglinski6564]

they won't be doing much coding in near future as they run out of money and energy. They are to expensive, to fault prone and make codebases a hell to debug. They are just chatbots.

[@WoodyJohnson-r3v]

​@@maciejglinski6564 Already 30% or more of the code on github is generated. As far as the money and energy, Microsoft is about to reopen a nuclear power plant just to power its data centers which also means they are not hurting for funds either.

[@sdstorm]

The L take is strong with this one.

[@ricdeh1701]

firrrrrrrrrrrrst

[@OfficialBeeswax]

Ask ChatGPT to list 10 English words where the third letter is 'm' and you will have an answer to the title

[@guyincognito1406]

This is choice right here. Because most of the words I thought of under those conditions do fit the bill for the answer. And then I asked GPT and got “aisle” lol and other words when I told it that they were in the 4th letter not the third it tried to gaslight me.

[@nachojimenez2420]

what is this bullshit remover? I need it.