Still hodling?
Still hodling?
Soon : teach yourself a CS degree using the AI tutor! Also don't forget to ask it to design a graduation certificate for your outstanding achievement!
The framework desktop is a bit over 2K USD and also offers 128GB unified memory, so for the price of the sparks you could almost get two of them, with about 256GB memory. I wonder hoe the spark compares with that?
the latest GPT models excel at tool from my understanding. Perhaps a database can act as memory and memory retrieval is a tool call to the database? It avoids doing prompt engineering and causing context rot?
@frame.work thank you for removing Google's ReCaptcha from your website, it goes a long way to respecting your costumers!
Love these technical writings. It is so freeing to read experts insights, unfiltered by PR or marketing.
Let's help preserve video games so they can be enjoyed by future generations rather than lost in the abyss of time.
If you're an EU citizen, sign this European Initiative, which is close to passing. It forces the EU Commission to take action.
eci.ec.europa.eu/045/public/#...
#StopKillingGames
Just passed 800k!
@doesitplay.org just passed 700k!
eci.ec.europa.eu/045/public/#...
@jameslee.bsky.social Hi! Do you have any more feedback to share on your move to Linux and how your workflow has possibly evolved since your video documenting your breakup with Adobe?
Did you ever get a reply? If not, no qualms!
Is there an embargo date for these reviews to be published? Or is it free for all?
It's a trust issue. And they haven't convinced me that they deserve it. To each their own of course.
That is interesting, I was unaware. I only saw the clause in ChatGPT saying that the outputs were owned by the user, hence OpenAI deflecting any responsibility. I may be misunderstanding their terms but it was still there as of February.
2) As much as you might want to trust the companies building LLMs, they built them disregarding copyright, so why would they not keep my code to use it for training despite claiming they don't train on it?
The article briefly covers copyright, but not in this way and I think it's a shame.
This article cover nothing about the two main reasons I don't use LLMs at work:
1) the risk is high that it produces a verbatim copy of someone else's code without letting me know where it came from because it memorized it. So high can I verify that I don't breach any copyright?
I'm not up to date so might have missed the news, but how's the job hunt going? Have you found a position you're satisfied with or are you still looking?
Great writeup, as always. Love these technical deep dives.
Regarding your website, is there a chance you can remove/replace Google's ReCaptcha? It is particularly user privacy invasive. Would love to have your input. Thanks!
Could you describe a use case for this? Is it meant to be used recursively to first request the implementation of multiply before asking something making use of it? Or am I missing the bigger picture perhaps? Thanks!
Isn't a recession defined by two consecutive declining quarters? It may already be declining this quarter but the numbers will come out in Q3 I guess
Aren't waymo cars also backed by remote safety drivers?
I don't know that it rebutes the authors claims, but I also don't know if cosine similarity is a good metric to use to evaluate. These embeddings, have they been constrained during training with a cosine similarity loss function? If not, how does the conclusion change for other metrics like L2-norm?
Or luck... The guy is unhinged.
Is this linked to the occasional sound hiccups which happen on resume from sleep?
Will you populate it with @frame.work Desktops, with 128GB hybrid memory each? Could be quite a powerful setup.
Any chance the full res images are available somewhere? :)
While you're in there, make sure to clean your membranes and joysticks, it's usually well worth it. Maybe replace the joysticks with some hall effect ones if necessary. Your controllers will feel like new!
I wonder if the SLOP acronym was chosen with the knowledge that it is also a term being used to describe content similar to spam but generated automatically using AI.
I'd love to see the same support for ROCm, it also needs some love (especially from AMD).
Or else wait for the @frame.work desktop with 128GB of unified memory for $2k, following all standards wherever possible. Shipping later this year.
It uses the latest Ryzen AI Max+ 395 (doing it from memory, don't quote me, it's a heck of a name!)
frame.work/desktop