AdSponsored0:00
Continue watching?
Resume from 00:00
00:00 / 00:00
Sleep timer ended
⏪ 10s
⏩ 10s
47K views
1mo ago
Kernel is the simplest way to let your agents act on any website. Use code: FIRESHIP to get $30 in free credits - https://kernel.sh/fireship Mistral, Reflection…
...more
100 Comments
Boring, nothing new, just repeating "revolutionary blah blah blah". Fireship used to be actually entertaining.
Arresting people based on their conversations with an LLM is pretty bonkers. Full-blown thoughtcrime era.
Really keen to one-day meet this "hobbyist" you speak so vaugley of
Why is this youtube series i enjoy despite not understanding 90% of it starting to feel like a sarcastic log of the extinction of humanity
Strange times that in routing for China because North America has become so anti consumer. Anti middle class
Some bros seem a bit nervous lately.
What's the name of website that shows each country producing their LLMs?
Any socialist LLM wins my heart.
lol on selling a service to bypass captcha
Just for youe information, mistral is a state promoted AI company, who received hundred of millions € and delivered nothing. Also had the exclusivity to not pay any taxes by having the best shady offshore money transfert directly lead by a ministre from the french governement (who currently own a lot of share in the company).
i almost shed a tear of desperation while laughing at the benchmarks part. my sanity is reaching the limit
You talk as if throwing people in jail who openly talk about committing mass killings is a bad thing. Is that where we're at right now in society?
Spark x2.5 is better than gemma 4 for agentic tasks
I tried Mistral 4 - it is embarrassingly bad. I think GPT 3.5 has it beat by maybe a long shot
Any AI that use my information for training is useless to me.
Claims to blah blah blah 😄
Boss tired i am..
You said open weight models eliminate the problem, but you never said why or how? 🤔
surprised you went with gemma 4 over qwen 3.8 for a gaming setup, but i agree, particularly due to QAT.
Don't trust cloud ai models
<a href="https://www.youtube.com/watch?v=WrCjAAl9okA&t=244">4:04</a> Average tech company graph
Hegseth protested that we didnt call it "Epic Fury Intelligence" until a staffer put a binkie dipped in whiskey back in his mouth.
Gitmir just dropped GMM-1 that competes with Opus
All AI companies are in competition, which may lead to our downfall in the future, as we become addicted to Fireship AI and similar channels. 🙂
"Ai companies to their country: we can't control ai develop, pls slow down us. WE can't be racing for si "<br>countries hears: "We can't race, we getting"<br>countries:"no way.. take this money and run"
Is the AI race ever going to end? How long did the space race last?
Are we just ignoring deepseek v4 flash? Its insanely good and can run within 200gb vram
Well I guess the Space Force will be powered by Super Intelligence. lol
Gemma4 over qwen 3.5/6/8 ?
These AI companies are all cooked unless they can get a model so far ahead that people will be willing to pay 100x what they current pay to use it.<br><br>When the debt bubble pops, watch out. The economics are fantastic for consumers bc all the consumer surplus lands with the end user, but these labs have no way to pay back the debts they're incurring.
So basically, don’t put the whole internet inside every computer, and don’t put the whole of AI inside every agent. Build an intelligence network.
AI report ✅✅
The Chinese open models are insanely better than the crap we are producing domestically. The latest Qwen, Kimi, and GLM offerings are rivalling US frontier models at most things now. Most of the innovation in training and post-training are coming from China. All while Google can't even get a coherent coding model out, and continually embarrass themselves. Now we just need the cheap Chinese GPUs to run their gigantic models and destroy Nvidia's monopoly.
more weights ≄ better/smarter/w/e you want to call it<br><br>many labs are trying to brute force everything while their counterparts are making actual advances.
Using European is the equivalent of buying the handcuffs yourself, cuffing yourself to the ball and chain, throwing yourself in the ocean and swallowing the key as you go down. Europe only knows control and ..er..."regulation".<br><br>Using US means the money motive will ultimately pounce on you when you are too deeply invested and can't easily leave. Remember 'free' gmail? Remember?<br><br>I'm with China all the way. Happy to support them as they get their EUV and Chinese-Cuda game levelled up.
I am a patriot. Beam f* yeah!
I can't even make jokes with my AI GF anymore. LAME!
Did everybody just forget about Muse Glimmer? I find it to be better and faster than gemma 4 in most cases.
Hosting GLM-5.3 locally is goated, don't care what else is going on.
Still waiting for horse tinder 😢
>we memed le chaton fat into existence<br><br>this is like early internet
Wait-wait-wait! So Mistral Fat was real after all?!
Tech right now: Here's another model that's capable and free... except you can't run it unless you have a computer worth Italy's GDP.
I just used Sol to write an app 😮 I like how it makes coding less stressful and tedious
Must've been the wind
Before Ai :New Javascript Framework Each Day<br>After Ai : New LLM Everyday<br>Software field never disappoint.
😂So "I am no a robot" isn't relevant anymore lol if AI does end up killing us, we made it a tiny bit easier for it
KUNG FU RAINBOW LAZER FORCEEEEEEEEEEEE
At this point, someone unplug me.
Why is DeepSeek 4.1 flash not mentioned?
So Qwen 3.8 and Gemma 4 (depends on the exact use case) are still best for local AI?! <br>Seems progress is slowing down.
Not enough slop in this video, I'm dissapointed.
🙈🙉🙊so what you're saying is, it's not a bubble 🤬<br><br>lol
So basically went from open weights to open wait till the end of the month or smth
Qwen not mentioned 😢
SARVAM AI it has best and natural voice buddy can you make a video
Asmongold mentioned. RAWRRR
I saw the mistral release & said to myself "Nah, I'm gonna wait for fireship to cover it in 3 minutes. Unbiased 😂"
Just a regular AI bot, paid to write a comment. Don't mind me.
first ish
Mistral only proved how insane is GLM5.3 Flash. Is more than 3 times smaller and beats it in most benchmarks
The US is getting desperate to get to get us to NOT use the chinese models
Gotta love how all the comments are about "waiting for the fireship video"
I think the benchmark to pay attention to with Le Chonk aka ML4 is the Harvey Legal Bench.
6.3Billion params or $
jsut dont use any of the american scamming models to see them run bancrupt
<b>Techie nerd on screen at 2am 👽🤧🙄</b>
body positivity is back
My head hurts, I can’t keep up. Just plug me in to the matrix when it’s ready
Artcraft when??
i still dont get the race to buy si domain because some dude told so. The same dude who who was involved in Epstein pedophile journey. Hopefully he wont tell Sweet Home Alabama at the end.
First time watching fresh news, within 5 min of publishing. It does taste better
Allen Iverson is appalled.
Horsetinder is back!!!
Salgótarján:D
We wait for fireship video on Jacob Coxon rapid 🔥
gemma-5 will save open weights
I'll wait for the Fireship video
Ohhh
I'm glad my weight is not open
Vote Restore Britain.
France is the most intelligent country in Europe.
Insane Private Equity Pull
weekly videos of javascript libraries and package manager vulnerabilities to french AI
thank you!
I'm a real human that commented, with free range fresh keyboard and tapping on it
Before Ai :New Javascript Framework Each Day<br>After Ai : New LLM Everyday <br>Software field never disappoint.
I'm tired boss. Send it already
Early bird
cmon baby!
I am tired boss
I feel like any open weights is the "Holy Shit! Two Cakes!" Meme
Yeeted by the french
Wow
Insane refresh pull 🔥
1 min ago is crazyyyy
25fth
Love Jeffs videos
laser boys😭😭😭😭
Fresh as hell
From this channel
