Thursday, March 23
Daily Tech News 23 March 2023
Triple Frog Edition
Triple Frog Edition
Top Story
- The SEC has issued what is called a Wells notice to crypto exchange Coinbase, warning of likely regulatory action against the company. (CNBC)
What the notice did not say is what the regulatory action might be for, leading Coinbase CEO Brian Armstrong to go on something of a tirade on Twitter. The company has been asking the SEC to provide regulatory guidelines for them to follow, but the SEC seems to think it is better to rule by fear than by, well, rules.
Having failed utterly to take action against FTX the SEC is now determined to put all the horses back into the barn and then set fire to it.
Tech News
- There's a looming replication crisis in AI research. (AI Snake Oil)
More specifically there's a looming replication crisis for any research that involves the products of ChatGPT creator OpenAI, which in reality is anything but open. OpenAI is shutting down access to its Codex AI, giving researchers three days notice before a hundred scientific papers were consigned to the reproducibility dustbin.
That site looks interesting; it throws cold water on a number of overheated subjects in the AI space.
- Nvidia's RTX 4000 SFF is a half-height Ada Lovelace professional graphics card. (Tom's Hardware)
Perfect if you need a second graphics card but your special edition Hololive PC case only has half-height slots after the first one.
It has 20GB of RAM and four mini-DisplayPort ports, delivers roughly the performance of the previous generation's RTX 3070, and uses just 70W of power. The 3070 itself has 8GB of RAM and uses 220W of power, so that's a pretty substantial improvement.
The price is, unfortunately, $1250. It would be quite a good card otherwise.
- Meanwhole Nvidia's H100 NVL has 188GB of RAM and fills four full-height PCIe slots. (AnandTech)
And uses around 800W of power.
Price is not even mentioned, but if you assume it will cost somewhere between a new car and a new house you won't be disappointed. If you wonder who is in the market for such a thing, Nvidia's marketing says it offers "12x the GPT3-175B inference throughput as a last-generation HGX A100".
Yeah, it's aimed squarely at OpenAI.
Disclaimer: No, b-e-e-tles. An inordinate fondness of beetles. Now turn that music down.
Posted by: Pixy Misa at
06:07 PM
| No Comments
| Add Comment
| Trackbacks (Suck)
Post contains 388 words, total size 3 kb.
50kb generated in CPU 0.0155, elapsed 0.1134 seconds.
56 queries taking 0.106 seconds, 346 records returned.
Powered by Minx 1.1.6c-pink.
56 queries taking 0.106 seconds, 346 records returned.
Powered by Minx 1.1.6c-pink.
