Something is wrong here. I have a 4070ti Super 16gb , a 7950X and 32gb of DDR5 6400 and 14b runs loads quicker . In task manager it says (for 8b) it uses 8gb of 16gb vram and with 14b it seems to use 11-13gb of vram and allocates an extra 16gb of actual ram of which i've not seen it use more than 0.2gb. Using latest Studio drivers and have CUDA toolkit installed. Edit, uploaded the Strawberry question on my own machine on my channel (might take a while to appear).

Install dual OS with Linux. Hopefully you get better result

That's very fast and that totally was a dad joke

I tried the 8B models for Llama 3.1 and DeepSeek. I wanted a calculator and the Llama did an exceptional calculator. Easy to use, it allowed me multiple calculations until I told it to stop while the deepseek took way longer, printed a code that didn't even work and I needed to tell it that is didn't work. This is exceptionally bad for such simple task that even 3B models probably even smaller can do a calculator. When it did the calculator it was hard to use with multiple steps and shut down the program after I got the answer. Really not great at all. The whole hype is around the big model that is roughly comparable to models much smaller than it.

it’s crazy to me watching that people actually seem to think china’s like gonna win the race now just from this one model that our latest unreleased one is already far past, headlines still have a huge grip on everyone including investors

mhh in my 3060 it runs at the same speed

anything less than the biggest model is a meme. the hype is about open source and quality parity. not about trying to run it on a crusty 4080.

Did you lose money on Nvidia stock?

I heard it cannot say anything bad about China or CCP. Can you maybe ask it something like whether China has ever oppressed its own people historically?

i ran the 32GB version using LM Studio. it was better than this, but i didn't love it. the full model thru the deepseek website is pretty solid however.

Please try LM Studio instead ollama.

It's funny the stock market thinks this will lower demand for GPU's while a 4080 doesn't even have the compute to count the r's in strawberry, not to mention training the model in the first place.

it works well and fast with the same model foe your two questions in China

My base model M4 Mac mini is doing these tasks way faster than your 4080..

They wanted to liquidate nvidia and needed an excuse

you realise there is a 671b model as-well?

If you were to do this project in 2025 would your choice still be the same?

Hey Chris. Meant to tell you RIP Bayside. Lost my pup last year as well and got a little cattle dog to carry on the flame. I’ve transitioned to cyber security (still in RVA) and have to put in that extra effort to make sure my man is getting all of his energy out lol. Stay safe brother.

A C guy here. I have to say this is a very excellent explanation. I feel like anyone talking about C is not doing enough disclaimers "I'm getting lost here/there.", but you put it excellently in just one simple "I'm not much of a C guy.", the best part is I don't even feel like it was needed at all. The IT field needs more people like you. That is "I don't have to understand EVERYTHING, but nor should I act like I know my tools are best for any job."

Linode is a shit show they started with Akamai shit heads.

Great for developers, bad for user browser!

Anyone here who still uses C/C++ for work? Among them those who wish they hadn't? I'll the the first

I have worked with several of this editors. Ckeditor, quil.. They are all not good. If its easy to implement, its lacking in some features. It it has the features it lacks documentation and has tonns of issues. This things are all a mess

ahhh

Bro just hit the "Apply All" button. Come on.

Hi Chris, so 4 years later how do you feel about nestjs, would you still recommend it and are you still running your site on it?

Chris, did u ever think about other career? What would be your number2 and numer3 picks after programming? Cheers (video please)

Python sucks.

Bro just press the damn apply button

Thank you

One more thing i noticed is that it outputs much better Ruby or JS code than Python, perhaps it is a function of what code it has read during training. And one thing is for sure Ruby devs have some nice Software engineering practices in the community compared to python. In python community shoving all the ugly code you have produced in a class is the limit of software engineering they know. So much of python code in the wild is just bunch of functions in a file, each function doing 5-8 things, as if no concept of SOLID exists in Python community.

It can go very fast and belt out code but doesnt think through all the variables. In which case you have to think through all the problems, have the forethought , then craft a well thought out prompt tehn it executes well on that. So a glorified Junior developer!

where AI has been very unhelpful / extra-time consuming are things such as deploying projects to multiple platforms (ElectronJS builds/deploys for example) or annoying Dev Ops type work such as intricate Azure issues

So far I haven't found Cursor worth it. Just using Claude or ChatGPT as a rubber ducky has been the most helpful. I already have vim set up like cursor if I want it but I almost never use it.

0:58 Yeah

Thank you for always bringing that sobriety and straightforwardness to things in the industry. It's especially useful for new people going into the field when getting filled with all sort of information and on top having douchebags like him around confusing more and giving a stumbling block on people .. 👍🏽💪🏽

Mattermost

While I think image/video AI generation is cool, it's definitely not even a fraction as cool as LLMs to me

Slack!!!!

Slack, hands down. Teams is an abomination, something that should've never existed.

my group uses Teams. it's ok. I do like the huge storage limits to create a channel and upload artifacts to it.

Teams sucks. Slack is great. But companies use Teams because it’s free

Teams is horrible

Discord is usable thing, Slack is.bulshit, Teams - in the same.bullshit.league

Actually both systems suck. I have to work with both and everyday I'm praying someone finally brings out a new system which is userfriendly, easy to understand, has a good search function for archived data, works offline if necessary, doesn't crash, has a fair price.

Teams sucks. Performs horribly, tons of bugs, disorganized chats... the list goes on.

Reading the title "...Without the Hype" and then hearing the video start "...we've all been waiting for." is a bad sign. I don't trust any company whose CEO has no idea what the SHIFT key does on his keyboard. Notice at 0:55 he quickly removes the horrible looking video so you don't get a good look at what a scam this is. Jellyfish don't run. This "real review" of Sora ends with "I'm not really sure what to think to be honest." You are killing me, bro.

I'm in stiches! "I asked for three little pigs, we got two." You're too patient with this nonsense, as if, we are two thirds of the way there, whereas this story is meaningless without all three. This is the most epic waste in the history of mankind. (Text much less so, it's a better google and i'm glad to pay for the Plus.)

Dude u are such a HATER lol. A year from now JUST LIKE WITH GPT, which u also trashed HARD at first, you'll be using an updated version of this probably fairly often.

Good demonstraition for why I actively avoid AI tools at least for now. I feel the overhyping and extreme use of these tools have overshadowed the limits we currently have and I struggle to find many use cases outside a short niche list such as stock footage and short promotional videos.