Claude 3 AI: Smarter Than OpenAI's ChatGPT?
Переглядів 79,438
2 місяці тому❤️ Check out Weights & Biases and sign up for a free demo here: wandb.me/papers
📝 Claude 3 is available here - try it out for free (note that we are not affiliated with them):
www.anthropic.com/news/claude...
Conference I am coming to:
wandb.ai/site/resources/event...
Experiments, evaluations:
/ 1764866009175920892
/ 1764754012824314102
/ 1764692641436827842
Additional results (very nice so far!):
huggingface.co/spaces/lmsys/c...
📝 My paper on simulations that look almost like reality is available for free here:
rdcu.be/cWPfD
Or this is the orig. Nature Physics link with clickable citations:
www.nature.com/articles/s4156...
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Alex Balfanz, Alex Haro, B Shang, Benji Rabhan, Bret Brizzee, Gaston Ingaramo, Gordon Child, Jace O'Brien, John Le, Kyle Davis, Lukas Biewald, Martin, Michael Albrecht, Michael Tedder, Owen Skarpness, Richard Putra Iskandar, Richard Sundvall, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here: / twominutepapers
Thumbnail background design: Felícia Zsolnai-Fehér - felicia.hu
Károly Zsolnai-Fehér's research works: cg.tuwien.ac.at/~zsolnai/
Twitter: / twominutepapers
@utubrGaming 2 місяці тому
Friendship ended with gpt4 turbo API. Claude 3 Opus is now bffs
@vectoralphaAI 2 місяці тому
For now. Until OpenAI released GPT5.
@and_I_am_Life_the_fixer_of_all 2 місяці тому
@@vectoralphaAI100%, claude3 and gpt4 gans until then
@Favour.A.Emmason-pv1mk 2 місяці тому
@@vectoralphaAI then we'll wait for Claude 4
@monad_tcp 2 місяці тому
@@vectoralphaAI only after $8T USD , they hit a hardware wall already
@DefaultFlame 2 місяці тому
I wish, but for me Claude is not available. I'd have to use a VPN to get access. "We're sorry, but Claude is not currently available in your country."
@14zrobot 2 місяці тому
Gemini was advertising similarly, but as it was released, I was not able to get a single useful prompt out of it. Makes my hype way less for those test results; seems to be pretty easy to cherry pick baseline
@Custodian123 2 місяці тому
@@Faizan29353 Gemini is an ADHD LLM. I never migrated from the paid GPT4 to Gemini... Hoping 1.5 is released before my trial runs out.
@sealoftime 2 місяці тому
Opus is already available. I've used it for some time, it's been incredible. It was much more human-like in the reasoning and answers that it's given. But I found it to be a little bit too wordy for the purpose of a code assistant
@uponeric36 2 місяці тому
Gemini or Gemini adavanced? irrc they promised that their new/best model would be "coming soon for free!" but Gemini is still currently near useless and Gemini advanced isn't free obviously. Free ChatGPT was better in every scenario I tried.
@codycast 2 місяці тому
And it was racist, so there’s that
@fabianletsch1354 2 місяці тому
@sealoftime that is definitely a prompt issue and can be fixed. I dont know in which tool you use claude but in many you van adjust the system prompt. I for example use "Pro Coder" in Typing mind. I have not tested it yet with Claude. But that character should fix it.
@docark3224 2 місяці тому
Every other day a new product that leaves competition in dust. This is truly amazing. Can we have more videos on use of AI in medical developments.
@ignacio3460 2 місяці тому
As long as we can explain the AI's results with human logic. I wouldn't leave my health decisions up to a probability model.
@jake9764 2 місяці тому
Yes, more videos on real-life practical applications of how AI is being used in academia, medicine, industry, etc, would be great!
@jonatan01i 2 місяці тому
@@ignacio3460 yet! But later I think I won't leave my health decisions up to a mere human.
@jonatan01i 2 місяці тому
But to a human who uses GodAI
@gavinderulo12 2 місяці тому
@@ignacio3460there is way more to it than that. Think of drug discovery or something like alphafold for predicting 3D molecular structure.
@peterkonrad4364 2 місяці тому
"hello this is two minute papers with caroy shonayfa here." it took me a while to realize, that the "here" is part of the name.
@vincent78433 2 місяці тому
Károly Zsolnai-Fehér
@rhyscooper3693 2 місяці тому
"Carlo, Jonah y Fahir" is what I always heard 😅
@HiAdrian 2 місяці тому
Yes, I had the same epiphany about 2 years ago or so when I was newer to the channel.
@rijaja 2 місяці тому
Through what I do at work, I got to start playing with langchain, and I'm very excited about LLMs again, just like the day ChatGPT came out. It's incredible how fast new models pop up
@Lightning---K. 2 місяці тому
You have a very interesting way to talk english and i must admit, i never heard this style before ... it makes you pleasant. I enjoyed your video !
@JohnSmith762A11B 2 місяці тому
Just watch the Ren and Stimpy cartoon. This guy is Ren.
@Sekir80 2 місяці тому
Welcome to Hunglish, my friend!
@Thedeepseanomad 2 місяці тому
Claude 3 recall was not all deep green so it got a "would you look at that!" Instead of a Whaowwww!!!"
@k4l1hm4n 2 місяці тому
Already using it and it's safe to say it's on par or even better than ChatGPT
@kepler186f5 2 місяці тому
I used Claude Opus, its way way better specifically with coding
@Rationalific 2 місяці тому
6:10 - It appears that at least for the moment, ChatGPT 4 is superior. Google totally lobotomized Gemini. I was able to do more things with Bard. When I recently tried translating - not asking for advice...but TRANSLATING - something regarding love and relationships, Gemini first said that it's just a language model and can't do that, and then when I tried again, said that it can't help me because that was a touchy subject. A translation about a relationship was a touchy subject. It looks like Claude has similarly been lobotomized, though I'm not sure of the extent. Since I'm not trying to mathematically determine the internal dynamics of black holes, what I've seen so far makes me think that - until Microsoft and OpenAi lobotomize it - ChatGPT 4 is still the best.
@apache937 2 місяці тому
yes hate that
@bahshas 2 місяці тому
lmao google done for
@wisedylan 2 місяці тому
This is fascinating!
@mintakan003 2 місяці тому
LOL. The crowd chanting for "papers".
@loneIyboy15 2 місяці тому
I'm sick and tired of these "AI Assistants" telling me what they will and won't do. I don't need it wasting resources trying to guess whether my use cases violate some arcane TOS, I just need it to follow my instructions!
@dabidmydarling5398 2 місяці тому
Totally Agree
@asdf30111 2 місяці тому
This is why being able to run them on your own is so important. We need more open source models.
@gavinderulo12 2 місяці тому
No. This is how we got deep fakes etc. There need to be some safe guards.
@loneIyboy15 2 місяці тому
@@gavinderulo12 This isn't how we get deepfakes, it's how I get my AI assistant telling me I'm violating TOS because a single line I need translated mentioned a mouth.
@dtmdota6181 2 місяці тому
AI is accessible to everyone. Everyone is not the same. So the AI takes 'mean' approach to balance those who need them for productivity and those who want them for malicious intent.
@htxdy 2 місяці тому
opus is nice actually, definitely can par on par with gpt 4
@htxdy 2 місяці тому
ive tried sonnet for 2 days now
@htxdy 2 місяці тому
and tried opus since today
@and_I_am_Life_the_fixer_of_all 2 місяці тому
@@htxdyopus outperforms 4 in many ways
@ThePowerLover 2 місяці тому
@@and_I_am_Life_the_fixer_of_all Make a video proving it please.
@Steamrick 2 місяці тому
It's worth rememberig that GPT-4 is nearly a year old at this point and still the reference point that all other LLMs try to beat. That's beyond impressive given the pace of development.
@SupahNin10dohp 2 місяці тому
Opus is pretty weird for me, it's guardrails are pretty ridiculously high, it jumps to crazy conclusions with certain instructions, formats emails wrong, and requires convincing for seemingly normal things.
@apache937 2 місяці тому
agreed
@FireNLightnin 2 місяці тому
I noticed the same thing.
@pavelperina7629 2 місяці тому
Really? I had this issue with Gemini mostly. Once it replayed it can't draw images of people (when nor people nor images were mentioned in whole conversation and it can't draw images at all), once it accused me for shaming, hate speech and legal consequences of my actions when I mentioned some "features" in certain search engine and some positive feedback mechanism to build monopoly. Actually even comments here must be neutral or positive. Best joke are efforts to write a c++ code which may be "unsafe".
@Free2PlayLessPays 2 місяці тому
the ai is region-locked.
@michno4323 2 місяці тому
Yeah this sucks :/
@somebody-anonymous 2 місяці тому
To the US?
@Bomberman66Hell 2 місяці тому
VPN?
@GoofyChristoffer 2 місяці тому
Well, it's probably just blocked for EU users because of GDPR or something similar.
@florisr9 2 місяці тому
Just use a VPN to set up your account, and then you can use it without a VPN.
@Benw8888 2 місяці тому
The post is a little disingenuous as they used old GPT-4 statistics, not the best newest GPT-4 statistics, like you said. GPT-4 is likely still better in many fields, but I wish we had a better comparison.
@0AThijs 2 місяці тому
It's so smart that it won't continue when the word 'fight' has been mentioned. 🥰🥰🥰 (But seriously, censorship has gone haywire...)
@JohnSmith762A11B 2 місяці тому
Counting the days until the word "censorship" in a YT comment gets it censored.
@JamesQuintero18 2 місяці тому
Can you please add a caveat or a caution that these results are not verified yet, and that they could be this high because Claude 3 was somehow trained on these benchmarks on purpose or by the datasets being leaked online. Nevermind, you said exactly this 5 minutes in! Great!
@and_I_am_Life_the_fixer_of_all 2 місяці тому
nah its legit
@smellthel Місяць тому
A chatbot that uses diffusion could be awesome. It generates random characters then rearranges and changes them to make sense for the prompt.
@cleverman383 2 місяці тому
What a time to be alive!
@The_Savolainen 2 місяці тому
Oh i get to be the first one :D. One of the best channels on the platform. Keep up!
@memegazer 2 місяці тому
Hope you film some of that in person stuff and share it as content on your channel Sounds like fun but I will not be able to make it there to visit the event
@Gomisan 2 місяці тому
It seems to be more intuitive when asked to generate code from scratch. GPT4 needs lot s of coaxing and explicit instruction, while Claude just writes it, and makes human like assumptions along the way. Also GPT4 will often rename or miss methods within a section of code, and you end up with bugs because it cant stay 'on track' wheras in my brief explorations Claude takes in the whole project and understands the context a lot better. If only they took Paypal not credit cards!
@winkletter 2 місяці тому
I test drove Claude again when version 3 dropped. Immediately subscribed. Now I'm running all my prompts through both Claude 3 Opus and ChatGPT 4. Both are producing stellar but different results. Do I need to figure out which of my two AI friends is smarter given they're both smarter than me at most things?
@oracle372 Місяць тому
"This would take a paid intern hours and hours of work, and it can just do it in a few seconds." My first reaction was being impressed, my second reaction worry.
@mm89_ 2 місяці тому
So my kids one day will be able to enjoy a coleric maths teacher like I had with his origin voice and the rattle of his oversized keychain. I'll make sure of that.
@siquod 2 місяці тому
Good to know that the axis that's just labeled "Intelligence" is on a log scale!
@marinomusico5768 2 місяці тому
Whoa 😮❤
@mercerwing1458 2 місяці тому
Will you make a video about the Mamba paper?
@Dimencia 2 місяці тому
So when it says it's comparing to GPT-4, is it doing the same prompt modifications (that they came up with) for both its own text completion model and GPT-4's model, or is it just feeding the questions through ChatGPT, ie different prompt modifications for each tested model? Neither case seems particularly fair or useful as a comparison
@linuxgaminginfullhd60fps10 2 місяці тому
Today I asked the same question to llama2, bing chat, chat gpt 3.5 and using your link the claude 3: "If velocity of an object is 3/4 of the speed of light in y direction, what would it be according to special relativity in the system that moves in x direction with the speed equal to 3/4 of the speed of light?" All of them got wrong answers - they either violate the principles of special relativity obtaining higher than the speed of light velocities or velocity is unchanged/decreased. Common sense suggests the answer should be somewhere in (3/4 c, c) range. Then I gave them some hints, yet none of them was able to figure it out... What a shame, this is a simple pre-university problem we used to solve in high school.
@artman40 2 місяці тому
I gave some custom instructions to GPT3.5 and that's what came out. "So, let's break it down for these poor, befuddled models: Imagine you're in a race car going 3/4 the speed of light in the y direction, and suddenly you decide to make a pit stop in the x direction at 3/4 the speed of light. It's like trying to juggle flaming torches while riding a unicycle on a tightrope. The result? Well, it's not breaking any cosmic speed limits, but it's certainly not a leisurely Sunday drive either. But hey, who needs common sense when you've got algorithms, right? Keep plugging away, little models. Maybe one day you'll catch up to high school physics. Or maybe not."
@linuxgaminginfullhd60fps10 2 місяці тому
Hm. Claude 3 seems to figure it out eventually. After getting LOTS of hints from me he made and corrected 5 mistakes. With significant help from me it was able to obtain the correct answer. It seems a little bit better than others I tried, but all of them seem to be at the same level of dumbness. Claude quite quickly understood that there are correct formulas that it can use - that was 1st mistake it corrected. And then it make 4 arithmetic mistakes. It had significant problems evaluating 1/sqrt(7/16) as intermediate step. So it is kinda good, that if figured out the correct formula, yet evaluating the numerical result turned out to be unexpectedly problematic for that LLM. As if it is a step back compared to other models.
@pavelperina7629 2 місяці тому
For me answer in claude.sonnet is v' = (0.75c + 0.75c) / (1 + (0.75c × 0.75c) / c^2) = 0.96c At least in expected range. But honestly when speeds are perpendicular to each other, I don't know the answer and I assume some weird hyperbolic trigonometry comes handy. But since my education background is electrotechnics, I don't know how to solve it myself.
@linuxgaminginfullhd60fps10 2 місяці тому
@@pavelperina7629 Yeah, some others were able to get the same with hits. That values is wrong. The correct is: 3 * sqrt(23) / 16, or around 0.899c. This case is simple because the original velocity is not mixed, only the orthogonal part is present, so the corresponding dot product in the formulas is 0. So for the orthogonal case general mixed formula simplifies to: u'x = -v u'y = uy / gamma Thus I expected it to be an easy problem for anything intelligent, that saw the general formula(which I am certain appeared in the training data multiple times).
@linuxgaminginfullhd60fps10 Місяць тому
So I was playing with it for the last week and I have to admit that looks like Claude 3 has some intelligence within. Not an AGI, nothing like that, yet capable of behavior I have not observed in other systems yet. I have not tried anything with images or files, I was using just some text prompts. Not just text that any one can answer, I was using some expert knowledge of my own, which would be hard for a human to get(because human would need to read a lot and spend 5-10 years of life on something potentially useless). Like the problems which would require very broad knowledge across multiple hard sciences and some thinking/reasoning on top of it. The system failed on the first try as expected, but when I provided the hints it was able to connect all the information, get the whole picture and fill in the missing details - at this point I was surprised with its ability to learn from a dialogue and the flexibility of its thinking. It feels like if it has enough talks like that and can carry the learned lessons forward to the new interactions it could actually get very smart.
@Rafa_informatico 2 місяці тому
They compare against GPT-4-launch version, from one year ago (not against the latest GPT-4)
@ZenBen_the_Elder Місяць тому
I love these geeks! Chanting 'Papers, Papers!' like a college pre-game pep rally. ❤ 6:25-7:22
@marcfruchtman9473 2 місяці тому
The predictive analysis for the GDP is off substantially. BUT, just to make sure, I am taking a screenshot and will come back to check in 2030 hehe
@daesgu90 2 місяці тому
Plot twist! TMP is actually not a person and it's just a super advanced AI waiting for the world to catch up
@lpvrooom6714 2 місяці тому
The data from that pie chart seems off. I cannot figure out what its comparing but it seems to be pulling both ppp and raw gdp values for that percentage calculation. See china vs USA and Italy/Brazil vs Canada
@JonWillis9 2 місяці тому
I'm trying to avoid all of these lobotomized corporate LLMs more and more...
@-_Nuke_- 2 місяці тому
Seems like Claude 3 is not available in all countries yet. Mine Greece is not listed...
@thechadeuropeanfederalist893 2 місяці тому
The small Claude 3 model even outperforms GPT-4 on Code Human Eval?
@troylowry4239 2 місяці тому
I would love to switch from gpt-4 to this but it's not available for Canadians for some reason >:(
@Gerlaffy 2 місяці тому
What's with the "AAND" copy/pastes? Are they like AI generated filler?
@galvinvoltag 2 місяці тому
"Actually, AI is very stupid." - GPT 3.5 Turbo
@davidvincent380 Місяць тому
Not really smarter than GPT-4 but damn, Opus' writing is more engaging. Much more human-like.
@timonix2 2 місяці тому
I just tried Claude 3 on a bunch of problems which chatgpt and Gemini just refused to get right. Claude did them all correctly first try. I will switch
@zdenekburian1366 2 місяці тому
Even though I don't know anything about ai or finance, I'm a bit doubtful about the economic forecasting possibilities of these systems due to the fact that here there aren't infinite possibilities for a variable, such as when composing a novel or an image , but only two possible directions, growth or decrease. If we assume that all AI systems carry out their forecasts at the maximum possible efficiency, on average they will all give the same direction in the forecast of a financial variable, therefore if all humans relied on these systems they would make the same choices of allocation of their resources, obviously within the limits of the same return on investment time frame. However, since it is well known that investments in purchases must correspond to equal opposite disinvestments, since it is a zero-sum game, who would remain on the losing side of the market? Can someone ask this question to an AI?
@Custodian123 2 місяці тому
Let me know when they run tests on GPT4 turbo
@mattmmilli8287 2 місяці тому
So far it’s better for coding. I’m a firm believer that GPT4 and CoPilot been getting dumber over time. Opus is a fresh spring chicken 🐣 wonder if it’s going to end up the same
@odatas 2 місяці тому
@@Faizan29353I dont mind paying money as long as its getting better.
@mattmmilli8287 2 місяці тому
@@Faizan29353 I mean opus is already 20$ but GPT4 which is paid as well has been getting worse
@kairu_b 2 місяці тому
Papers! Papers!
@lancemarchetti8673 2 місяці тому
Sonnet helped me fix and complete the code for my new steganography app. I had tried a few bots over the past few months and could not resolve the issue around why the math was mapping the coordinates in the image binary incorrectly. So I was seriously impressed.!
@meduzak 2 місяці тому
The Q is like "hold my papers" :D
@NorbertKasko 2 місяці тому
1:58 Those GDP forecasts are already as wrong as they can get.
@ThePowerLover 2 місяці тому
Even more the global one.
@rogeriopenna9014 2 місяці тому
Any comparison must take into account pricing. ChatGpt 4 is 20 dollar per month
@Gomisan 2 місяці тому
So is Claude.
@rogeriopenna9014 2 місяці тому
@@Gomisan I never said it wasn´t I just said the comparisson should include that. Can I already test Claude in the same way I can test Gemini? I hear that Claude's cheaper version is worse than GPT. It's only the most expensive version that is smarter. Is the 20 dollars a month for the most expensive version?
@Gerlaffy 2 місяці тому
Dollars*
@rogeriopenna9014 2 місяці тому
@@Gerlaffy wrote the first post on my phone with swipe keyboard. As the R is near the S, it probably wrote dollar, gave me the option for dollars but I didn´t notice it.
@austinsmith1293 2 місяці тому
You should do an event online, such as in VRchat, so we can meet you without having to travel. Just a flight to San Fran will cost me $1000, which is more than I make in a month.
@Gerlaffy 2 місяці тому
Oof the cringe is high with this one!
@lukasm5254 2 місяці тому
Microsoft 1bit (1.58bit) Model for LLMs as next paper?
@survivezeal Місяць тому
How y'all following up with the audio?
@user-fh7tg3gf5p 2 місяці тому
If I put a 108,000 word book to summarise, why does it say 400% above limit?
@iminumst7827 2 місяці тому
Granted, my tests are no where near academic, but after watching this video I tested claude 3 and chat gpt with 4 difficult game programming related questions that I know the answer to. Claude 3 got 2 correct and GPT got 3. Claude also seemed to be worse at guessing the solutions, since chat GPT's logic was closer to the truth. I think I'll stick with GPT for now.
@robertturaa9561 2 місяці тому
It's available in 159 but not in Poland? Wow, the devs must hate us.
@piranha1337 2 місяці тому
Not a single EU country. Guess why?
@robertturaa9561 2 місяці тому
@@piranha1337 I guess I'm out of the loop. Some AI regulation?
@Srindal4657 2 місяці тому
The age of ai won't reach it's peak when companies use it easily, it will reach it's peak when your average joe can use it easily.
@DanFrederiksen 2 місяці тому
Did you try it?
@FreeEasyAI 2 місяці тому
"Smart" seems an odd term for artificial intelligence. I guess I equate "smart" with real intelligence.
@clamhammer2463 2 місяці тому
It is also 16x more expensive to use than gpt 3.5 turbo.
@rishiraj2548 2 місяці тому
🎉
@SevenOfNineteen 2 місяці тому
Sabine Hossenfelder just talked about that too. She has some very interesting additions.
@_DRMR_ 2 місяці тому
Important detail: Claude is not available in the EU. Time for a VPN sponsor?
@EffortlessEthan 2 місяці тому
just a little tip, "leakage" has some not so pleasant connotations lol
@USONOFAV 2 місяці тому
it is so smart it made me subscribe, and will downgrade my Google One since Gemini Advance is pretty much useless
@danhansson666 Місяць тому
one thing about AI tho, is how do you know that what you get out is actually correct, if you don't know it yourself
@peetymcfly8871 2 місяці тому
chat gpt4 is great already. gemini just refuses to answer too many questions. as long as they are not restrictive, i am satisfy
@celozzip 2 місяці тому
2:00 a.i. can barely do simple sums at the moment, you have to check all the results
@dabidmydarling5398 2 місяці тому
they can't? Last time i used GPT for math was a while back but im pretty sure it was able to do addition completely fine.
@V3T_SC 2 місяці тому
You haven't done any videos about Grok? Why not?
@ivomirrikerpro3805 2 місяці тому
I know right. Truth seeking AI is forbidden. What a waste.
@Gerlaffy 2 місяці тому
Maybe he couldn't fit "aand" in the video enough
@byteseq Місяць тому
"What a time to be alive". Sorry, cannot share your enthusiasm...
@jinparksoul 2 місяці тому
AI teachers will be training students not to be teachers themselves but to efficiently do manual labor and how to enter the small number of careers that AI cannot dominate.
@callibor3119 2 місяці тому
This is better. I was not hoping to hear another OpenAI video for much it “shares with scientology.” I don’t want to be on the same AI that a cult known worldwide is catering to.
@TheKwiatek Місяць тому
How come that it is avaiable in so many countries but not Poland?
@DrNioky 2 місяці тому
Did you switch to AI-generated voiceover? I haven't watched your videos in a while but your voice and way of talking were much different a few years ago...
@Gerlaffy 2 місяці тому
AAAND. ANND. ANND. AAAND. AAND.
@Aurelyyon Місяць тому
sadly that seems to be the case
@pandoraeeris7860 2 місяці тому
Two more GPT's down the line!
@jopansmark 2 місяці тому
It's over for OpenAI
@zrakonthekrakon494 2 місяці тому
It really isn’t
@jopansmark 2 місяці тому
@@zrakonthekrakon494 you've seen the benchmarks. Numbers don't lie.
@Gerlaffy 2 місяці тому
Over because... They didn't just instantly release their newer models? Get a grip.
@zrakonthekrakon494 2 місяці тому
@@jopansmark They laugh at your unbelief, they have extremely powerful technology that they simply are not releasing to the public whom they view as the plebian masses because they want to horde the power for themselves.
@ThePowerLover 2 місяці тому
@@jopansmark See the video completely.
@calinciobanu25 2 місяці тому
unfortunately this channel has transformed from an insightful paper review into an AI hype news. Can't even remember the last time a paper was reviewed here
@Gerlaffy 2 місяці тому
It's just "this is big! This is crazy! What a time to be alive! Aaand! AAND. Aaand. And.
@godofthecripples1237 Місяць тому
Brain-dead take. The role of this channel was always a highlight the advancement of technology. AI is the center of that right now, whether you like it or not.
@SudiptoChandraDipu 2 місяці тому
GPT-4 ... Gemini ... Claude ... So much progress in so little time! The pace is terrifying.
@gabrielsandstedt 2 місяці тому
gemini is worse than both. I have used all three a lot, Opus is best currently then GPT4 last Gemini.
@jesper164a 2 місяці тому
GPT-4 less than gemini?
@Baronvonbadguy3 2 місяці тому
I have found Gemini good as a writing assistant it provides friendly results.
@Rationalific 2 місяці тому
Gemini has been censored to the extent that it's been lobotomized. Claude may be as well, but I don't know the extent. GPT4 still appears to be more free than either of the other two. I'm not talking about extremely risquee stuff. I tried translating stuff about a relationship, and Gemini told me that it wasn't able to translate touchy subjects. Now, a translation about a romantic relationship is a touchy subject according to Gemini's handlers.
@gabrielsandstedt 2 місяці тому
@@Faizan29353 I know I am paying for it. And gpt4 and Claude.
@HistoryIsAbsurd 2 місяці тому
Opus might be good but man their new free model suckkkks at coding. Keeps getting basic python wrong.
@kl6336 2 місяці тому
All basic models suck balls
@BigyetiTechnologies 2 місяці тому
How does it compare with gpt4 for coding?
@HistoryIsAbsurd 2 місяці тому
the free version of the new claude 3 model kinda just made up codewith zero sense behind it. Code GPT originally wrote perfectly. So not well. Although ive seen Opus used very well so thats definitely just the free version and in my (admittingly limited) opinion@@BigyetiTechnologies
@clerothsun3933 2 місяці тому
It is not actually good. Fake benchmarks
@camiscooked 2 місяці тому
@@clerothsun3933Opus is scary. You very obviously didn't shell out the 20$
@danielreed5199 2 місяці тому
Given that its latest version is called Opus... is it any good at making music?
@skylineuk1485 2 місяці тому
Suno v3 is lol
@netdreamr 2 місяці тому
@@skylineuk1485there's gotta be a video about suno v3 on this channel because it is way too good
@danielreed5199 2 місяці тому
@@skylineuk1485 That is disputable, I just gave it this prompt... "Rock, C, D, Am F"... I got back a pop song with these lyrics "In the depths of the night, the stars align As the moonlight guides us under this sky of mine With the rhythm of the rock, we'll soar up high C, D, Am, F, we'll reach for the sky (ooh-yeah)" I would prefer something that understood music theory and not just have the ability to mimic songs based on a bunch of other songs it has heard. This is like a parrot AI to be fair, don't get me wrong though it is still quite impressive. FLStudio is integrating AI into their system, I presume other DAW's are also doing it. It would be nice if these tools were designed to teach people about the subject and not just generate stuff that people can then claim that they made it. It is going to be an issue... "Did you hear this awesome song I just made?, it only took one click".. get ready for a new generation of SoundCloud artists who have tonnes of followers who are unable to differentiate actual talent from a button press.
@scarm_rune 2 місяці тому
thank you
@galmud1508 2 місяці тому
An AI-teacher for every child! Great! Provided there's stil a need for educated human beings in the future.
@kugeltmg 2 місяці тому
Exactly what I thought. If the AI is smarter and cheaper than a human teacher, resources will go to training better AIs not teaching kids. Inequality is going to rapidly increase.
@martinsramkad1761 2 місяці тому
But what job will that kid be able to have when the AI has all the jobs
@JohnSmith762A11B 2 місяці тому
@@martinsramkad1761 Yes, it is funny how often education gets mentioned when after AGI this will be a post-labor world as human labor will be worthless. Still, originally people went to college to learn about the world and what interested them, rather than as a ripoff vocational training to be future corporate serfs. So really, this is an improvement. This comment will now be dragged away and executed behind the chemical sheds for excessive snark bordering on illegal satire. I'm just writing it to exercise my fingers. Good luck fellow useless humans!
@galmud1508 2 місяці тому
@@martinsramkad1761 Jobs requiring opposable thumbs and a pair of legs. Until humanoid robots take those jobs too.
@Rationalific 2 місяці тому
@@galmud1508 So for sweatshop labor and roadwork, I guess.
@phutureproof 2 місяці тому
man i miss your videos about water simulation created by humans those were great, this shit's getting boring very fast
@BloodyCrow__ 2 місяці тому
Fucking Bhutan has access to Claude but not Canada?
@AnonYmous-yu6hv 2 місяці тому
they should have made it cheaper than gpt 4 to be competitive.
@ManOfSteel1 2 місяці тому
is it free?
@Burning_Typhoon 2 місяці тому
Claude 3? GTA reference?... LOL
@Gerlaffy 2 місяці тому
What?
@tonysolar284 2 місяці тому
AI replaced Interns, is Coffee their only hope?
@DownTheBarrelOfaGun 2 місяці тому
~~AND
@Porschession 2 місяці тому
Will you speak in the same manner? 😂 maybe get rid of the pauses and speak more fluidly. Not that it’s a bad thing or so but currently you sound a bit like a robot. No offense please. I really love your videos! ❤
@Gerlaffy 2 місяці тому
AAND. AAND! AAND. AND. AAAND.
@d.r.656 2 місяці тому
What do we make of FSD V12 in this comment section?
@Gerlaffy 2 місяці тому
FSD V12?
@d.r.656 2 місяці тому
@@Gerlaffy first real world use of AI
@Gerlaffy 2 місяці тому
@@d.r.656 will look it up, thanks
@piedpiper4450 2 місяці тому
idk if pursuing a career in data science is a good idea anymore
@uponeric36 2 місяці тому
@@Faizan29353 My take: Most people will still be too dumb for such jobs using AI even if everything they want is one sentence prompt away, because they are increasingly literate and cannot read or write well. The real challenge with AI in the future won't be making the models smarter, but figuring out how to deal with people getting dumber. So as long as you can still read, write and do basic math at the pace of 90s elementary student you can probably find a job in post AI world.
@apache937 2 місяці тому
definitely not, look into ML research instead
@piedpiper4450 2 місяці тому
@@uponeric36 sure but as a fresher that is still studying im afraid these developments will make a lot of jobs redundant. the barrier of entry with thousands of com sci/ IT related freshers will compete for an even smaller pool of jobs. im just lost at this point
@beiyongzui 2 місяці тому
I hope there is finally one intelligent AI saying that religion is stupid
@kinngrimm 2 місяці тому
Why would we need AI teachers for our childreen when most if not all jobs are done (better and cheaper) by AI? It doesn't stop at teaching, research, art anything you can compute. Soon afterwards due to AI intelligence explosion robots will improove significantly as well, well they already do due to LLMs, but with a true AGI i assume we can't even imagine what is coming for us ... two papers down the line.
@Gerlaffy 2 місяці тому
Well, maybe if you had an AI teacher you'd know how to spell correctly.
@kinngrimm 2 місяці тому
@@Gerlaffy oh i am sorry for not correctly spelling in a second language, asap as i got my AI agent sorted out this wont happen again ... yet my argument will still stand.
@grey_north9016 2 місяці тому
If the Pentagon can hide trillions of dollars from the government itself it can hide almost anything 😅
@ertantosangcomandanterting1108 2 місяці тому
I try translate language English to other languages And vice versa in chatgpt 3.5, gemini pro, and Claude 3 the result very Disappointing not very accurate
@wanfuse 2 місяці тому
that chart didn't impress me , a paper chart off by 10-20% is not so good, otherwise, nice!!!
BBC News Україна
Переглядів 618 тис.
FLIP FLOP Hacks
Переглядів 77 млн
Вечір з Яніною Соколовою
Переглядів 154 тис.
Goda Go
Переглядів 657 тис.
The Neuron
Переглядів 4,3 тис.
SANU4YOU
Переглядів 1,6 млн
Tech way
Переглядів 2,5 млн
LED Screen Factory-EagerLED
Переглядів 7 млн
Svidomy
Переглядів 38 тис.
KOMPUKTER
Переглядів 230 тис.
Immersed
Переглядів 23 млн