This is fantastic work, really impressive is an understatement. I really hope this sets a new DeepSeek-esque standard and starts another the death knell for companies continuing to cosplay as frontier labs (like Cohere).
This was something I was genuinely excited to try while in Bangkok, took a Grab across town to make it happen, but it was it is honestly not good, the Google reviews seem to coincide with the consensus that while the concept is cool, the execution isn't and it tastes very meh.
I do wonder if there’s some pluralistic ignorance going on, where travelers convince themselves it must be amazing because everyone else seems to think so.
It didn't give off the vibe of years of collected flavors; it was a thin broth and it didn't taste like much else other than beef broth from a fancy instant noodle packet and a ton of MSG (and to be clear, I'm normally a proponent of MSG but it genuinely was overdone here).
Maybe it really is just the volume that they're going through that's affecting the taste and composition, because they were doing decent business, but this was the biggest disappointment on my short visit to Bangkok.
If I had to reckon, it's because the web comes in very many shapes, and outsourcing that work to a generalist LLM/SLM like GPT Nano is expensive, and doing it deterministically will never catch all the edge cases as well as a purpose-built encoder when run at webscale.
It is scraped from Pixit. They sell lost/found, evidence + seized item management systems. [1] The listings are public; it was cool OP turned this into a mini art piece.
This is so cool. Thanks for doing this. I was able to get the Optiplex 7050 with the i5-6500T and 8GB of RAM (no SSD) for $40 USD about 2 years ago, shocked it's $100 USD now! I brought 8 for some reason, this makes me feel better that I at least purchased it during the glut.
Please correct me if you have contradicting data but: Neuralwatt's price per token vs price for energy comparison doesn't seem to take into account the cost savings from cache hits that other providers offer on pure token rates. The comparison seems to assume every input token is a cache miss.
On top of that, the cloud offering doesn't seem that well-run, they randomly blocked a colleague's API key for a couple days without any heads up, had a weird rate limiting bug and they have been deprecating models without redirects with very short notice, all while taking weeks to onboard new models. I assume some of these problems would be addressed if we had an SLA/enterprise contract.
It's a promising idea though. They offer a $5 trial credit (with an aggressive rate limit) though so no harm in trying it out.
>And then consider if you are are the spouse or parent of a murder victim or missing person, and are innocent. You are almost certainly going to be a potential suspect, but you also want to help them find the actual criminal, and if hiring a lawyer makes you "look guilty", then the police might focus their investigation on you rather than finding the real perpetrator.
I agree. This is a real problem in law enforcement. The issue is that while they're competing interests, my right to remain innocent should rationally supersede the deep profound desire for justice (but I recognize it rarely plays out this way).
>Looks like they have very effective collaboration with DeepSeek and Kimi.
The collaboration is informal. People don’t seem to realize this, but the Chinese internet for programmers and developers today feels a lot like StackExchange in its heyday. There’s a huge emphasis on sharing knowledge, because sharing what you know builds your profile, and becoming a rockstar in a subfield is one of the only ways to get ahead.
Competition in China is ruthless. But unlike in North America, where individuals are often bound by agreement to hoard knowledge because it can give them a competitive edge, the competitive advantage in China is building face and peer recognition. And that comes from proving that you are worthy of being a "master/teacher", and that extends to the valuation of your knowledge business. For example, the third wave coffee shops in China, the master roaster is often called "master/teacher" once they win a roasting competition and start sharing new knowledge of roasting in the public sphere, and that's a title of sincere respect.
You can see parallels with those that apply to give talks at conferences and post snazzy technical presentations they give in the US, but the bar for what qualifies as new knowledge is far higher in China because there's a massive ecosystem of people rushing to outcompete what you have to offer, and once the ball gets rolling on knowledge sharing, lots of people will go off and build upon that knowledge or try to build businesses on top of that, which in turn produces more knowledge.
Reading developer forums in China, once you crack the code (I find Gemini will get you a good chunk of the way with good translations), they are really quite far ahead with what they're willing to share. And I suspect in great part, the decision to release open-weights is heavily tied to that concept of building face/peer recognition = building valuations.
Sadly not all of them, much to my ill discovery! They are fun souvenirs and I encourage everyone to collect them if they want a free momento (ideally bring your own stamp pad).
.I urge everyone reading to unsubscribe instantaneously from the NYTimes for their business practices. Do not do business with unethical companies.
You are not wrong for thinking that, but I encourage people to consider that generally the business and editorial areas are largely independent of each other because of the value of editorial independence in case they think that the lack of ethics applies to their journalism too.
Hard to say. People made the same prediction a year ago because we supposedly ran out of training data. There could be indefinite rapid compounding improvements so long as there's free money out there.