aespa(x)
@aespa__init
Followers
71
Following
792
Media
99
Statuses
877
A certain scientific GPU accelerator. Work in tech but not technical. LLM enjoyer. Math challenged.
Canada
Joined February 2024
Of course that's your contention. You are a lowbie account; you just got finished reading some greentext post, Zephyr's probably. You're gonna be convinced of that untill next month when you get to Ahmad's GPU threads. Then you're going to be talking about how you will go into
0
0
2
I have $10 left on my Claude credits, might as well burn it on Opus 4.5 and see if it's worth considering a Max plan.
1
0
1
Was watching a video on $10K M3 Mac Studio and wife gave me a look. It's over for me, boys.
0
0
0
“If you owe the bank $100 that's your problem. If you owe the bank $100 million, that's the bank's problem.”
HSBC forecasts that OpenAI is going to have nearly a half trillion in operating losses until 2030 https://t.co/RUZmj0qx51
0
0
1
Wonder how many people will miss the part Karpathy says “pass 1 is manual”.
I’m starting to get into a habit of reading everything (blogs, articles, book chapters,…) with LLMs. Usually pass 1 is manual, then pass 2 “explain/summarize”, pass 3 Q&A. I usually end up with a better/deeper understanding than if I moved on. Growing to among top use cases. On
0
0
0
For 2) you have to ask how well it actually works, because it sounds just like AMD’s HIPIFY. Is MSFT better at converting Cuda to ROCm than AMD themselves? Even if one grants this, I find it hard to believe it works out of the box and is performant. Just because the code
I read this script and was left with a few unresolved questions. If anyone knows the answers, please leave a reply to this tweet. 1. Google is having this kind of success with TPUs; what about AMZN’s Trainium? 2. MSFT has developed a tool that converts CUDA workloads to
0
0
1
It’s @shakoistsLog.
🚨 Lana Del Rey admits she has a secret X/Twitter where she rage-baits people.
1
0
5
Too many back and forth “yea”, “right”, “exactly” to try to simulate natural conversation, and then always ending with a one line summarizing thought. People don’t talk like this and I certainly don’t want to hear a complex idea being interrupted by “uh huh”.
0
0
0
I don’t know if it’s just me but the NotebookLLM podcast voices/tones are too droning and robotic for me to listen to.
1
0
0
The dOOMed hahaha.
0
0
0
Half the agents at my work is just connecting documentation to a LLM.
0
0
0
Every time I see “it/that/this matters” in an email from someone, I want to tune out.
0
0
0
I see that article about AI killing consultants, and while their rates may come down, consultants serve a few critical functions: 1) take on some blame when the project goes south, 2) sometimes internal teams just don't like/want/care to collaborate or share information with each
0
0
0
Most revealing part of the Dwarkesh pod with Karpathy was that Dwarkesh doesn’t know what is Starfleet academy.
0
0
0