Excited to see this here! Only yesterday I used it to create an interactive native-like notification my coding agents use to ping me when they push a PR. I had tried a few different libraries over the past month but only hs had the ability to render custom panes over maximized windows.
Will be using it for more automation tools moving forward.
The problem is that dealing with regulators takes years and millions of dollars, reducing competition and societal benefit. He's quoting $200m in additional health costs borne mostly by Medicare/Medicaid. Regulations aren't a useful part of the system if they're gunked up.
I don't believe we have any indication that the big offerings (claude.ai, Gemini, operator, tasks, canvas, chatgpt) use multiple models in one call (other than for different modalities like having Gemini create an image). It seems to actually be very difficult technically and I'm curious as to why.
I wonder how much of an impact our being still so early in the productization phase of this all is. Like it takes a ton of work and training and coordination to get multiple models synced up into an offering and I think the companies are still optimizing for getting new ideas out there rather truly optimizing them.
Ooh, what are these ASICs you're talking about? My understanding was that we'll see AMD/Nvidia gpus continue to be pushed and very competitive as well as have new system architectures like cerebras or grok. I haven't heard about new compute platforms framed as ASICs.
Yeah I found that interesting. Flubbed embargo times makes sense, but could it also be that they're letting the news organizations have first dibs to build a little goodwill with the industry?
I'm not finding any direct sources from OpenAI, but here's this snippet from a Reuters article [1]
> Priced at 15 cents per million input tokens and 60 cents per million output tokens, the GPT-4o mini is more than 60% cheaper than GPT-3.5 Turbo, OpenAI said.
It currently outperforms the GPT-4 model on chat preferences and scored 82% on Massive Multitask Language Understanding (MMLU), OpenAI said.
...
> The GPT-4o mini model's score compared with 77.9% for Google's Gemini Flash and 73.8% for Anthropic's Claude Haiku, according to OpenAI.
For some more context: We don't know the size of 4o-mini but Mistral's just released NeMo 12B scores 68% on the MMLU. [2]
Just tried Toucan and it can't be disabled on localhost, a major pain for using it during work as an engineer. For those that haven't used toucan, it's an extension that translates words/phrases inline on a page with various levels of replacement frequency and complexity based on your proficiency with the language.
A YouTuber named DirtyTesla has done a series of drives over the years throughout his hometown where he's tracked the intervention and disengagement rates per mile. He often shows those numbers at the end of his videos.
"EGS are engineered reservoirs, created beneath the surface, where there is hot rock but limited pathways through which fluid can flow. During EGS development, the injection of fluid into the hot rock enhances the size and connectivity of fluid pathways by re-opening fractures. Once completed, EGS function just as natural geothermal systems do: fluids circulating through the hot rock carry energy to the surface through wells, driving turbines and generating electricity. EGS could provide up to 100+ GWe of economically viable capacity in the United States. This potential could supply green electricity to over 100,000,000 American homes, and represents a domestic energy source that is clean, reliable, flexible and renewable."
"Until today, people could enter another person’s phone number or email address into Facebook search to help find them. ... malicious actors have also abused these features to scrape public profile information by submitting phone numbers or email addresses they already have through search and account recovery. Given the scale and sophistication of the activity we’ve seen, we believe most people on Facebook could have had their public profile scraped in this way."
Hm. This seems like an interesting tidbit, I would love to know more. It seems to imply that many profiles have already been scraped in this way. A phone number is a really strong cross-domain identifier as we use it across a bunch of different online services. Collate your Facebook scrape with a couple data brokers and you've got a real strong profile of someone.
Will be using it for more automation tools moving forward.