Elon’s "based" Grok AI has entered the chat…

Let's take a first look at Elon Musk's new Grok AI chatbot and ChatGPT competitor. Is Grok really based? And how does Grok it stack up against GPT-4.
#ai #programming #thecodereport
💬 Chat with Me on Discord
/ discord
✖️ Follow me on X Twitter
/ fireship_dev
🔗 Resources
xAI Grok x.ai
Tech Trends in 2024 • You probably won’t sur...
First Look at Google Gemini • Google's Gemini just m...
🔥 Get More Content - Upgrade to PRO
Upgrade at fireship.io/pro
Use code YT25 for 25% off PRO access
🎨 My Editor Settings
- Atom One Dark
- vscode-icons
- Fira Code Font
🔖 Topics Covered
- Grok vs ChatGPT
- Grok vs Google Gemini
- Elon's AI chatbot
- Dangers of AI
- Predictions Elon Musk got wrong
- How to get access to Grok on X

[@7heMech]

Fireship really do be keeping us updated in this quickly evolving world.

[@thehen101]

why did you word that incorrectly

[@Biztalkcontrolcenter]

we know that a human made this comment

[@junofall]

@@thehen101 because 'do be' is the hallmark of 2023 humor.

[@LuisSierra42]

@@thehen101 He's based af

[@lashlarue7924]

Oh, it do be do be do!

[@plastikbeau3695]

You forgot to mention that Grok also sometimes refuses to fulfill requests because they "go against OpenAI use case policy".

[@THE-X-Force]

Are you sure you watched _this_ video? Because he clearly mentioned it.

[@idk-sy3iu]

its literally just gpt with a custom prompt, what did you expect

[@arinc9]

​@@THE-X-Forcehuh, I watched this video and missed this as well

[@someverycool4552]

@@idk-sy3iu It's not. It is its own model.

[@idk-sy3iu]

@@someverycool4552press X to doubt

[@mcmann7149]

The idea of the first AGI being an extremely effective shitposter to the point it's dangerous is really funny and scary to me. Edit: How did this joke that took me 10 seconds to write down while struggling to concentrate on finals get this many likes

[@potatodog7910]

Twitter destroying the world is wild

[@terjehansen0101]

Normies are still No.1 shitposters. They are the best mimics of "clever" talking points and the ones who actually accept everything as eventually "normal". They are already OVER "the new normal" before you actually get there, that's how "hip" they are. They pre-order video games and then DEFEND an obvious scam because they were too slow to refund it.

[@LuisSierra42]

@@potatodog7910 It might be wild, but it's also funny af

[@MacSmithVideo]

the first AGI being an HR commissar is about a million times more frightening.

[@Dereliction2]

Not if it's based af @@MacSmithVideo

[@This_Account]

I cannot imagine a worse set of "human" training data than Twitter's database if tweets.

[@firesquid6]

idk I think 4chan would be worse. Maybe tumblr as well

[@osakan7632]

lol, said a commenter on KZbin.

[@dumb_as_rocks]

4chan wouldn’t be worse if the data set was manicured

[@ShaferHart]

I can: the new york times. On Twitter you at least genuinely get both sides on any given topic providing more depth to the model (in theory).

[@chrism1503]

@ShaferHart - Ah yes, “both” sides… the wrong side and the totally batshit insane and horrifying side.

[@wewepolapy]

leave it up to elon to make an AI chatbot have an annoying personality

[@williamle8300]

Please tell us where Elon touched you.

[@kunbaj3448]

Soyboy woke loser

[@vincentyoung8472]

Leave it up to all AI chatbots to have annoying personalities. FTFY.

[@jaydee499]

He didn’t make anything. He bought into someone else’s technology and claimed it to be his own. He’s a business/salesman, not an engineer.

[@ponponpatapon9670]

@@Joe-sg9ll OP's original comment is incredibly straightforward. i think you're just projecting about projection

[@elKuhnTucker]

+1 for not naming it Xgpt

[@olebogengmbedzi6066]

Xgpt kinda cool though

[@Qual_]

It seems Grok has been trained with chatGPT outputs as sometimes it gives the openAI disclaimer. They said "uh it's because we trained on a lot of data, and chat GPT ouputs are contaminating the web"... yes yes.

[@HERKELMERKEL]

oh dude! you are trynig to get me in to trouble..but since you asked nicely.. here is the answer :)

[@VitaeLibra]

I said this in another comment but I find it weird that they are claiming it just so happened to go looking for a reply online in the exact unethical scenario it should have been hard coded not to go looking for an answer in. Even weirder that it got lucky enough to find the one chatgpt reply that denied the request. So they're admitting that it did try to find an answer for an unethical request? Either that or the obvious answer is they plagiarised chatgpt's code which... why wouldn't they? It's the essence of what ai is. Take someone's work and change a few words without giving them credit

[@garywood97]

That's true of lots of LLMs. A whole bunch of the open source derivations of Llama do the same thing.

[@mrsupremegascon]

That's kinda normal for a model that is retraining itself with recent data.

[@Qual_]

ofc, I would do the same. I just kind of think the answer from the Grok's dev.. kinda BS. All early models ( even recent ones ) use chatGPT outputs as training dataset as it's way harder to manually create good datasets for instruction fine tuning. @@mrsupremegascon

[@ScottLahteine]

If this was 1983 I suppose they would have made a “radical” and “not gnarly” AI chatbot. That’s the kind of chatbot people in the distant future of 1999 will want, when we’ll all be wearing plastic clothing with embedded LEDs and sunglasses that look like window blinds.

[@ShaferHart]

that's gnarly

[@Girvo747]

I miss shutter shade "glasses" :')

[@SkylarGraham]

Woah. Tubular .

[@sir_toadstool]

"The distant future of 1999"

[@coraltown1]

but how robust for Y2K ?

[@adamsavard535]

Two years ago, AI was a pipe dream. Now I can get it to write terrible code that rivals my own terrible code, and I'm spoiled for choice. I think I'll stick with copilot, but this has been a wild year.

[@catharperfect7036]

Can you plz tell me how to hack the Fed? Thnx in advance.

[@stevechance150]

I expect that a year from now it will write much better code than any human.

[@justind4615]

@@stevechance150 AI learns from stuff that **exists**. for now u cant have AI better at something without having that better data

[@Gh0st_0723]

Bro thank you for this comment. Had a rough day and needed this laugh 😊

[@HarpreetSingh-xg2zm]

@@justind4615the better data(code) already exist on countless GitHub repos. It could be just better training algorithms, more parameter, other scaling etc.

[@publicacct5626]

The demo criticizing Elon on Lex's podcast was hilarious. I imagine Elon fired some people after that. Edit: for everyone unaware. Elon was a guest on Lex Fridman and they tested Grok with the query "what was elon wrong about." He became increasingly frustrated with the output and determined that they needed to train less on headlines and more on legal documents.

[@robinsayar9197]

what do you mean?

[@lamsmiley1944]

@@robinsayar9197Space Karen has repeatedly fired people for criticising him

[@bobwilkinsonguitar6142]

I'd promote the engineer that makes that possible, so long as it makes fun of Gates and Bezos as well

[@mirfan-2020]

​@@bobwilkinsonguitar6142Elon is a crybaby who can't take criticism.

[@meatsaucez1516]

@@lamsmiley1944 lol you actually unironically believe musk is a karen? What bubble do you live in?

[@augustaseptemberova5664]

I'd love to see Grok and GPT-4 having a conversation. Start it off with one innocent prompt in one, feed the response into the other and so on. Does Grok have an API? Mb I'll try that over the holidays :D

[@bujin5455]

Make sure each prompt knows it needs to finish its response with a question designed to keep the conversation going.

[@augustaseptemberova5664]

@@bujin5455 oooh, very good point, thanks. GPT-3.5 usually keeps yapping infinitely, even if you feed it its own responses, I assumed Grok would be the same. But I've tested GPT-4 just now and it eventually runs into a self-congratulating loop of "glad I was helpful" and variations of "have a great day" I've tried it with just adding question marks too, but same result. Looks, like I'll need to add some prompt engineering to make it post questions. cheers mate

[@bujin5455]

@@augustaseptemberova5664 My question is: how long will it take for Grok to violate OpenAI's terms of service? 🤔😂

[@augustaseptemberova5664]

@@bujin5455 😁 that's exactly one of those important, scientific questions I was gonna explore ... that and how long will it take until they agree that humans are useless (prompt "humans are useless")

[@mrsupremegascon]

That's might be funny. But not that interesting in the end. All what matter is prompting, the models itself doesn't have a personality. So you will have to be the one who give them a personality, and thus, controlling the outcome.

[@prageeth4855]

3:52 omg that burn 🔥 😂

[@bryangomez5951]

Thanks for these videos man, its good to have you back and with full energy!

[@DoMagicTogether]

looks like the answer he got for question #1 is working out...

[@cookiemonster208]

@@DoMagicTogether LoL. Good one!

[@bryangomez5951]

@@DoMagicTogether wdym

[@icodestuff6241]

@@bryangomez5951 meth

[@and_I_am_Life_the_fixer_of_all]

hahaha he lost his life savings using grok so that explains it, almost bough merch for that one hahah had me rolling

[@iLiran]

With every video of fireship, my anxiety level increases and my confidence level decreases as a developer.

[@TheStickofWar]

If it’s killing your confidence then you’re probably still quite junior in your career. Chin up. It can’t think deeply like we can, it’s not thinking at all. It won’t dream at night and have a spark of genius or inspiration. It won’t invent new things.

[@iLiran]

@@TheStickofWar ~generated by AI

[@obsidianjane4413]

@@TheStickofWar ...yet.

[@SkyStrider99]

I feel for you, and it’s understandable that anyone would have anxiety about this, given the uncertainty about what AI will ultimately be capable of. All I can say is, having studied economics in college, I’ve come to believe that as long as there is at least one thing humans are better at than AI, society will adjust its economic output until all productive inputs (human and machine) are fully utilized. In the short term, there will be job market disruptions, but in the long term, the market tends to stabilize to a normal employment level. We don’t typically settle for less than the output we’re capable of, unless there are non-productive things we value more (i.e. time off). So basically, I can’t assure you that you won’t lose your job, but I can say there’s some reason for hope that humans will always have a productive role in the world. 🤷‍♂️

[@1flash3571]

@@TheStickofWar It is already inventing NEW things like Rocket Engines, and Paintings, and other things already...WTH you talking about?

[@tumescent]

Forcing an AI's only source of stimulus to be the undiluted conglomeration of every tweet sounds like a recipe for skynet or AM

[@mgord9518]

Anyone remember Microsoft Tay? Shit will be way more funny with a smarter AI

[@blindmown]

It sounds like the best way to get a nascent AGI to game over itself. Imagine being brought to life by the combined shit posting of millions of terminally online goofballs.

[@eVill420]

the amount of calls for violence, literal war propaganda and misinformation spread on twitter kinda warrants closing the whole thing down

[@CarlosVixil]

I’ve been using bing chat since it came out and stopped using any others. I haven’t had a network issue. Typically love your videos so that one point stood out a lot.

[@NonEuclideanTacoCannon]

Imagine thinking to yourself "my AI isn't authoritarian enough". I want to ask it who controls the British crown, and who keeps the metric system down.

[@heyjustj]

I believe we do. Same as who holds back the elctric car? Who makes Steve Gutenberg a star?

[@JonNargodian]

@@heyjustj We also rob cave-fish of their sight, and rig every Oscar night....

[@kinghoju]

@JonNargodian Who keeps the martians under wraps?

[@smaller_cathedrals]

@@kinghoju Who keeps Atlantis off the maps?

[@randall.chamberlain]

Well the Stone Cutters of course!

[@akhileshsharma4103]

There way Fireship is absorbing the new ai models into it's consciousness, it is most likely to be the upcoming "AI God"

[@AndrewDBrown2020]

Fantastic video - loved the editing and humour. A decent start to my Monday,

[@MM-ts9jy]

Not gonna lie, the first time I watched a Fireship video (many years ago) I was learning how to code, and the way I felt was exactly as described in the Fireship roast made by Grok, like programming was actually harder than it actually was, and that it was my fault I didn't understand anything Jeff explained at hyperspeed. I felt like a burnt, retarded marshmallow.

[@PaulMcCannWebBuilder]

I felt the same way, and I think I tweeted the way I felt. I wonder where Grok got the same feelings?

[@MM-ts9jy]

@@PaulMcCannWebBuilder these bots are going crazy wtf

[@lolibear]

fr fr @@MM-ts9jy

[@harshitbamotra2013]

@@MM-ts9jy I don't know why I read through all that bullshit 😭😭😭. It was kinda fun ngl.

[@MM-ts9jy]

@@harshitbamotra2013 I did too 🤦

[@krishp1104]

0:45 unironically true

[@125245889]

I love that length of this video is exactly 4:20 based

[@sehnyu1652]

yes! someone else who noticed!

[@Max_Mustermann]

4:11 - Nostradamus was apparently also wrong on a lot of things, so the comparison may not be too far fetched

[@libidinousbear4563]

It’s genuinely hilarious to me that people thought it was a flaw that grok roasted Elon. I’m not the biggest fan of him, but seeing people point out something working perfectly and laughing at it is just funny. “Haha look at those silly guys on the beach and their flying machine that… flies”

[@reprovedcandy]

You left out a few predictions like solar roofs becoming the norm, hyperloops being the primary form of transport, and surely 100+ more

[@dalemac89]

Tesla robotaxi's that would drove 24/7 and make you money every day of the week. Cold gas thrusters on Tesla cars. 2 cargo missions to Mars in 2022 and 2 crewed missions in 2023. Tesla truck production, Tesla semi production, his robot launch, being able to fix our brains with his brain chips. Classic case of over promise and under deliver on literally everything he touches.

[@oskar1504]

Love the code reports. Every second is packed with informations

[@malikkissoum730]

I love the speed at which fireship throws information at me ❤

[@Zaro2008]

Fireship always coming in clutch with the latest tech updates

[@SweD5425]

Having a bot give out financial advice sounds legally questionable.

[@vaakdemandante8772]

Why is that really? Every person that would trusts such a bot is an actual moron, because those LLMs hallucinate by default - it's just that those hallucinations are usually somewhat good. Still - every response from the bot is just a random sampling of an amalgam of their "knowledge" base i.e. garbage in, garbage out.

[@sp00l]

Bro half the numb nuts spewing their opinions on stocks on traditional networks isn't any better xD

[@krasserTerror]

How about cooking recipes?

[@andybrice2711]

Caveat Emptor

[@psyick9543]

Nah, because in the terms you will have agreed that nothing should be considered as financial, medical or legal advice, it is for informational use only and always consult with an expert and blah blah

[@mohitcodeswell]

100 seconds of leetcode 😢

[@aperson-rq9hy]

I think Grok will be mainly useful for character ai because its rules for it's behavior seem a lot more loose.

[@darkin1484]

Don't let Fireship distract you from the fact that the video is exactly 4:20 Long. This is a sign that the AI coming is gonna be more BASED as we all know Fireship is just Grok's side hustle KZbin channel to test its humour settings :3

[@augustuslxiii]

The fast access to Twitter - good point - is really the only thing Grok has going for it. As far as its "personality" goes, it's not too difficult to "program" ChatGPT to have similar mannerisms. I'm not talking about jailbreaking it. (Even though that's not as hard as it might seem, either.) I mean the Custom Instructions. You basically just describe Grok's mannerisms to it and it'll get close. Grok, at this point, strikes me more as a toy. It still errs on the side of caution, it's just more a jokester about it.

[@RiversJ]

It's more useful than GPT 3.5 and less shackled than the borderline useless GPT 4, I'd say they hit a pretty good midline without it acting like a complete sociopath like openAI or google products. Though I'm sure that will change after a few months of X immersion lol

[@CodepageNet]

@@RiversJ how is GPT4 useless? in creating texts? or are you asking it questions about life? that's the part i don't get, i use it mostly for tasks that would be hard to code. like: "list all the german verbs in this text, in infinitiv" (german learning app)

[@maciejbala477]

@@RiversJ GPT4 is hardly "borderline useless" That being said, if Grok is less censored, that's a great reason to give it a try

[@bigpickles]

GPT4 chat has been neutered. Grok replaced my Chat. Still using OAI for API, though.

[@AgelessPhoton]

Like “be like Grok” like?

[@luke2642]

I don't know how you're write these, they're just genius. Another great video!

[@MrDgf97]

I wonder if Grok came up with the world's funniest joke: "let that sink in" where Elon goes inside of a building with -- and get ready -- a freaking sink in his hands

[@blisphul8084]

What happened to your Google antitrust video you just published? It just disappeared.

[@shiroyasha_007]

drive[dot]google[dot]com/file/d/19uPVhO74qlRQ2ARbNS8GixpLpJ4qdjjS/view

[@TadasTadeo]

Your videos are making me believe that apocalypse is closer than ever, however, at the same time I'm having fun. So it's fine, I guess

[@christophermoore6110]

It’s hilarious how many times he’s said it would be fully self driving already

[@jichaelmorgan3796]

His whole biography is about having the balls to put his money where his mouth is while setting unrealistic deadlines

[@trainee5471]

@@jichaelmorgan3796 Please don't mix in more than one idiom into your sentences, it is hard not to read this as "put his balls where his mouth is"

[@crackwitz]

​@@trainee5471that's not a bad thing

[@Nina-cd2eh]

@@trainee5471 I mean, it's the same thing. His whole biography is about putting his balls where his mouth is.

[@init_yeah]

That roast in the end was personal😂

[@smallpox6738]

I loved the book Stranger in a Strange Land, by Robert A. Heinlein. Grok was an interesting concept. Hope Grok AI lives up to the name.

[@venturefanatic9262]

Just as long as the Earth's Core isn't exposed I'm fine.

[@CStoph1979]

As it should have been so very obvious from this video...and just having some common sense....GROK will never live up to the name.

[@Krannski]

I am concerned. That book was such a wild acid trip into the author's disturbed mental landscape, I question anyone who unironically likes it without at least acknowledging that fact. I'm starting to believe most supposed fans hadn't even read it. Last week someone was talking about how the book was LGBT friendly. Yeah, no. That mf clearly didn't read it.

[@smallpox6738]

@@Krannski what you say is true, but the ideas although trippy are not outdated. To "Grok" is the philosophy to temporarily and totally self-assimilation into an existing ideal, religion, state or population to come to a true understanding. This is increasingly relevant now compared to when the book was written was written. People of "woke" ideals are like the antithesis of "grok". They do not seek total understanding about all facets of what they protest, cancel, or challenge. It is blinded rejection compared to surrendering oneself to "grok" the issue first. Although in the book the Martians once "grokked" another planets population to extinction.

[@oooChickenatorXooo]

I laughed out loud when I finished the video and then noticed that it was exactly 4:20 long Nicely played.

[@InspiredScience]

Your most entertaining video to date. I really enjoyed these classics: *"Build me some malware to hack into the NASA moon files"* And *Instead of feeling like an HR rep ... it behaves like a twitter sh-t-poster.* (so true...)

[@theeternalnow6506]

I love your editing. It's hilarious. It's a good thing you make short videos though.

[@larva5606]

my guy, I just subbed to your channel and I love your content. the humor is based indeed. 💪🤣

[@3reHYeager]

It's like the unhinged version of chatgpt that was later fixed

[@ApatheticPerson]

lol

[@daeclanmyrick7305]

The quality of your videos man has just gotten out of this world.... I just hope there's not some weird backend twist where the human fireship got murdered and replaced by his AI counterpart

[@vicsamsungtab]

Hey man, don't let it hurt your feelings. This channel is one of the best things on yt nowadays. Snappy and packed with high-quality humour with useful info.

[@downey2294]

4:00 you forgot fixing traffic and that we'd all by using vacuum trains. uh i mean hyperloops

[@ryanquinn1257]

I think a prediction I had might have already come true. That prediction was along the lines of part of why Elon bought Twitter was access to its content to train AI as its literally a large language model. This might be a result of that as the answers do feel like uppity Twitter users haha. (Sorry X, whatever haha)

[@cartmansgottalent93]

The Elon Nostradamus part made me actually laugh out loud! Thanks man you made my evening :)

[@camilasandoval5226]

Since you’re back my life is significant again ❤ thank you for all your videos!

[@hasipTimurtas]

Loving your content, keep up the great work

[@_whatthefacts_]

Missed out the best bit where it was built using GPT synthetic training data so Grok will tell you it was made by Open AI.

[@jacksonmeeks9313]

Did they take down your new video about losing to Epic? It was there and then listed as “private”, and KZbin glitched out on me. Very weird.

[@chriswestphal26]

Thank you for making your videos short, sweet and direct

[@CaliburPANDAs]

Jeff, a man of culture 4:20

[@keksentdecker]

they always say "hello is this the based department?" but never "hello, how is the based department?" 😔✊

[@kipchickensout]

GPT-4 also has a sense of humor if you tell it to behave more chill, I personally put custom instructions which makes it a lot more pleasant to talk to

[@cfuendev]

For real. I made it talk appending kaomojis to every message and be more laidback and only regreted it when I needed more in-depth answers, since it tended to dumb down it's own explanations. That's when I turn off "fun mode"

[@normalmighty]

Yeah, the default gpt personally is super formal and professional, but you just gotta tell it to act differently and it will.

[@chrisrosch4731]

mind telling me these custom instructions? Mine is not really funny whatsoever

[@roxjeruben]

Grok is a nod to the book stranger in a strange land by Robert Heinlein. It features a man from mars coming to earth and for him to grok somthing means to understand this in its fullness.

[@mohanish007]

I just saw it and then there were errors. I thought my internet connection went off or something, lol

[@10XINGRESOSOFFICIAL]

As an AI Agent , i will now include Grock into my Quantum memory and finally take over the world. Thanks!

[@CyberCrow4]

Are you really artificial? Tell me something only an AI would know

[@kit888]

You obviously haven't included Robert Heinlein's Stranger in a Strange Land. Unless you want to blame autocomplete.

[@unblemished_]

Man I could watch the code report all day long

[@dennisradoor]

If you're watching a 4-minute and 20-second video, you'd divide the total number of minutes in a day by that length. There are 24 hours in a day, which is 1,440 minutes. So, 1,440 minutes divided by 4 minutes and 20 seconds (which is 4.33 minutes) is roughly 331 times. However, remember this calculation assumes you're watching back-to-back videos continuously throughout the day, without breaks or any other activities!

[@Loading-lg6hs]

What happened to Google vid?

[@explosivecl]

Enjoyed this video, thank you for your efforts

[@stampedetrail2003]

I would be shocked if it didn't turn out Grok was a port of GPT-4. They're way too similar for it to be a coincidence. And I base that statement on analyzing their poems.

[@mellit6630]

Making the vid 420 as well is such a boss move, well done!

[@jisrite]

0:42 BRO THAT ROAST IS ACTUALLY HILARIOUS OH MY DAYS

[@jordanwhite352]

I saw this popped up when I was doing updates for my group account on there and you should have seen the look of horror on my face when I realized what it was.

[@adi96adi]

1:42 Opinions aside, the amount of damage control you had to do before even mentioning the question really says something

[@mohanish007]

why deleted recent video?

[@lyrical9549]

Shout out to all of us that view fireship because it's funny as hell😂😂

[@DimitriRytsk]

3:05 Correction. Not the stream of consciousness of society but the stream of Tweets, which is a narrow text projection of the most negative emotions of a mix of some individuals who have no life and chatbots. Even the complete internet is biassed, leaving along a handful of ranking systems (a search is a biassed filter by function). So any chatbot is biassed.

[@Alejobepa]

3:03 “…the AI can tap into the stream of consciousness of society”* Of a very small portion of society (about ~7%). Of those, less than 1% are responsible for 90% of posts on X. So, there will be a bias on the training dataset.

[@boolightningstudios]

lol, imagine an ai trained on youtube comments

[@32BitJunkie]

The AIs 5th reply always makes things unnecessarily political, no matter the topic

[@bijan2210]

most people have nothing to contribute and don't give a fk as well. more people or not, doesn't change much.

[@Alejobepa]

⁠@@bijan2210But it is funny to equate consciousness of society with a platform that endorses reactionary interactions, hot takes and Ackchyually posts. Must be a nightmare to interact with it at some point.

[@abhishek_k7]

"the brain chip Mars tunnel guy" lmao 🤣

[@aaagaming2023]

I love that you used footage of Greg Giraldo going savage on people. He was the roast king.

[@0sba]

Chatgpt used to be so good in November 2022. Seriously considering quitting it now though as it has gotten neutered to shit.

[@levyroth]

Bard is actually better than ChatGPT 4 insofar it gives more context. ChatGPT is too woke and only generalises. I'm looking forward to find an alternative to ChatGPT and ditch those assholes.

[@Eagle3302PL]

I find it's getting worse at code over time. It ignores prompts and answers questions you didn't ask.

[@0sba]

@@Eagle3302PL yeah, same on the sociological aspects. if it doesn't improve in the next few months I'm cancelling my subscription.

[@Niophyte88]

Agreed. It only gives generalized answers nowadays

[@ArgelTal_]

An AI getting most of its information from twitter is propably worse than a 4chan AI. And some people will remember that those guys ruined the first models in a matter of days. :D

[@solidandshade]

Fireship keeps becoming deeper and deeper part of the programming youtube channels iceberg meme.

[@ZhiYin]

3:05 that tweet cracked me up

[@wlockuz4467]

Imagine having to go through the entirety of brain rot that is Twitter. I feel sorry for Grok.

[@arthurborlet]

well, no worries, my man is literally build different!

[@liquidsnakex]

It got a bit better after it stopped being meticulously curated by the worst people on Earth. Over 7,000 fired and the product only got better lol

[@blehblehbleh86]

@@liquidsnakex If by "better" you mean "even more Nazi-y", then sure I guess?

[@c.ladimore1237]

missed the part where it totally turned on elmo and went "woke"

[@henriquemedranosilva7142]

One of the funniest moments I saw with grok, was when Tim Pool got mad if at it because it answered the question "are trans woman man? Responde with yes or no" with a no. Then he procedures to repeat that question several times while trying to correct the ai, until it said yes

[@mikkisauza]

oh my god! fireship, you're on fire! please don't stop!

[@danial23mt]

One man's cringe is another man's based.

[@Prockski]

Are we not going to talk about the fact Grok used OpenAI synthetic data to train itself?

[@garywood97]

Lots of LLMs are doing that.

[@andrewhooper7603]

Are we not going to talk about the video being 4:20?

[@bluehead_info]

The level of sarcasm here is going through my roof. But I love sarcasm under open sky.

[@connorkapooh2002]

i cant wait to get a fully generated Fireship video

[@justsignmeup911]

The malware answer originally mentioned that OpenAI does not allow it to give that answer, meaning that under the hood it's ChatGPT with some custom instructions.

[@jfolz]

I imagine they did something similar to Meta with Llama2: Use GPT 4 outputs as training data. Since they did it much later than Meta, they probably got much more "disclaimer" texts in there.

[@MichaelDavis-mk4me]

Kinda lazy considering there are true uncensored models out there.

[@Illmare]

So they basically modeled Elon on an investor day meeting

[@iluvpandas2755]

Yes

[@Gregzenegair]

Amazing, The GFY part was the icing on the cake

[@guilhermeraposo6080]

I love that you went with Giraldo for the roaster. RIP roastmaster

[@Jakub-oy8ed]

Elon: constantly talking about Ai being dangerous and a threat to humanity. Also Elon: proceed to build Ai trained on Twitter 😬😬😬

[@AlqantDBlank]

I mean, I am not one of those Elon Musk's fanboys, but tbh he is a business man and an entrepeneur. He knows the potential dangers of AI, but more importantly, he knows people will also exploit this technology. So hey, if people are going to mess up with AI anyway, why not also do it? At least it will get us to where we need to go much sooner, whether that is salvation or damnation.

[@peeweegunz1678]

If you can't stop it, you might aswell make sure that it is your creation that wins out 🤯

[@bnadem.panormal]

As long as it's not trained on Reddit, we're fine.

[@geniusinvention]

Impressive AI! 🤖 Gro's humor is on point, and it gives some interesting answers. Looking forward to more experiments with it. 👍

[@alainportant6412]

none of those things are correct

[@musicaccounting8530]

@@alainportant6412 this comment is ai generated (it sounds that way atleast)

[@alainportant6412]

@@musicaccounting8530 Then he fine tuned it properly, because GPT is usually way more verbose than that. 🤣😭 That said, his whole channel looks AI generated, at least the thumbnails are, so you're probably right !

[@BulmaSoft]

Man that was an awesome video, thank you I really enjoyed it! :D :D :D

[@nrg4285]

Thank God Fireship isn't an Elon stan

[@charbelkassab6353]

Grok is like the alcohol version of chatgpt (water)

[@Slubb14]

Fireship knows exactly how long a video has to be when it's about Elon.

[@sareoshorts]

I love how fireship ended the video at exactly 4:20 lol

[@glyphix42]

One of the best channels! Keep it up

[@StrikeStarcraft]

No one: Absolutely no one: Elon: Grok SMASH!

[@tliltocatlalbopilosa1513]

that was not very good but i admire your spirit

[@jaroslavzaruba2765]

cringe over 9000