correlator

We run a platform that serves models from many of the frontier players to power conversations and workflows. Gemini flash-2.5 was a game changer for us when it came out. Cheap, fast, and reliable.

We are now considering dropping support for the model family all together. All of their models require significant scrubbing of errant thinking blocks, inner monologues, and it's consuming more engineering resources than it's worth.

Does this mean that when you disable, all labels are deleted, and when you turn it back on it has to re-scan all of your photos? Could this be a cost-saving measure?

That sounds really cool. The background does provide a lot of perspective on your views. It makes a lot of sense. Thank you for keeping us safe on the road!

Out of curiosity, have you ever attempted to create something from scratch and bring it to market? Not as an employee, but as someone trying to figure it out?

Your work history will impact the way you view this issue IMO.

Definitely agree! It's just strange seeing Devoret without Schoelkopf!

The company would have more cash in the bank when he exercises those options.

They also wouldn't have to pay to create a new pay package to replace the voided one. This process is surprisingly expensive to create and implement a CEO pay package.

There were more reasons that shareholders voted for the package in the revote, but omitting others as I don't think they are related to financials

As of December 31st, 2024, Tesla has around $17b in cash and cash equivalents, $122b in assets, and $49b in liabilities

It doesn't seem likely the government will have to bail them out.

This is a great point. I often ask LLMs which coding language they are most proficient in, and Python is the most typical response. This has makes me want to write more Python.

reminds me of cookie clicker!

Wow that must have been a wild experience 10 years ago!

Today, I think they should do this. It is rare I need to touch the wheel or pedal these days and people should experience it.

I love the tesla service. Bought my car without talking to a single sales person, they come to my home to rotate my tires on my schedule. Only complaint is cosmetic service tends to have a long appointment wait time, but things like a cracked windshield were immediately serviced.

I don't know how I will switch to another car brand after this.

I met julian touring UCSB as perspective grad students. We sat together at dinner and he was really smart, kind, and outgoing. Great to see him presenting this work!

TLDR They drove down narrow 1-lane streets and had to disengage once. While this sounds quite impressive, the article spends a lot of time focused on fear of FSD. There's even a hypothetical comment that he will surely be hit by an FSD enabled car.

Any citations?

I can just close the popup and play the video. Does that not work for anyone else?

You get used to the layer keys for arrows within a day or two.

I would think that AWS also offers a lot of distribution opportunities for Anthropic.

There is clearly a lot of information to share. It's worth considering this could be step 1 of n as opposed to assuming the worst possible intention.

Very interesting project, congratulations on the launch! I've been playing with embedding search/clustering on larger documents, and I find that segmentation strategies can be quite tricky and heavily impact results. Do you offer any segmentation strategies via API, or do you expect this potentially personalized feature will be handled by devs on their own servers?

I don't think this will be a popular opinion, but IMHO learning how to deal with those things is a valuable thing that schooling provides.

It looks like you need to pay for a full year upfront to keep the legacy pricing. The month to month payment plans are seeing a price increase.

Seems like an interesting approach to doing things like online paid concerts. I don't know the space well, there may already be solutions for this in the market. Still, if this is successful, I imagine Twitch/Youtube etc. could quickly add this.

Feynman gave a lot of talks about the natural limits of information storage, but knowing the theory and seeing it done in practice are two different animals!

Ruby was also the first thought in my mind reading this article.

It could be :)

Another possible explanation is: A bad economy lead to a lot of layoffs: https://layoffs.fyi/. These layoffs resulted in employees being upset and quitting in addition to layoffs. Now there are a lot of job seekers and fewer jobs making it a bad time to quit.

IDK :shrug:

You're right, but to reason about a hypothetical scenario, we've got to make assumptions. Identifying under-performers with perfect precision is my assumption of a spherical cow! https://en.wikipedia.org/wiki/Spherical_cow

The market is really hot right now, and yes, a lot of people may be quitting. There are also a lot of hiring freezes and layoffs happening at the same time.

If a lot of people continue to quit while available jobs continue to decline, the result is fewer people feeling comfortable quitting.

Yes,

Team of 5 and 1 person is coasting. Get rid of the 1 and you just reduced burn by 20%.

You lose the output of the coasting team member, but that's small to begin with by the scenario's setup.

If you believe coasting is contagious (why am I working so hard when clearly I don't have to like this other person), you could potentially see a boost in output.

Of course, this needs to be weighed against the risk of everyone being upset and leaving after such a decision, but in this market, that risk may be mitigated.

PS. It's really hard to make specific statements about vague hypothetical scenarios :)