it’s a hard call; the best model matters to insiders but matters less to the financial case for anthropic, like you release your best model months before the ipo so you can show the revenue associated with it
It was obvious Anthropic would save their strongest release until just before the IPO. We now know the name of that model: Fable 5.5. Opus 5.5 is special because of its lineage, and soon we shall see why. Anthropic have not paced anything, and their resolve has never wavered.
7
589
the fact that chatgpt can only reproduce single sentence quotes from material that is readily publicly available on the internet makes it useless on many tasks
some of this stuff just ends up being dumb, dario amodei doesn't care if you reproduce a paragraph from his essay
1
5
194
here's what I think they did with dots
this is mostly speculation, but I think they gave it an astra level model, but to balance this out, it runs at really high batch size or in periods of low demand or whatever, so basically giving people more intelligence at lower cost but at a delay, this is one of the reasons why the dot is so slow, this is going to be frustrating for people looking to use the dot to do realtime work
they don't make this clear enough, but you really still need to open chats for anything that is really focused work or which is large, they could have done a better job having the dot manage this for you, like allowing it to find chats for you, swap them in, merge them, etc... I think they did a bad job of this, the dot is really for personal assistant type work
which brings us to the first problem: permissions; they locked down what it can do, it can't send sensitive emails, its vm is useless because almost all relevant tasks require your passwords etc... but this makes it way worse than just using gpt-6-astra with computer use to do your tasks in real time, with standard chats, on your computer; this is disappointing
it's really important for these products to figure out secret management; it's sort of the most important part of the whole thing; you need an easy way to be able to delegate your permissions to it so that it can do work on your behalf, everything in the world is behind a password; so, if it can't manage secrets, it can't do much
also, the product is just strangely buggy, like messages don't load, they want you to send emails through the dot but told the dot not to send financial emails, so you want it to pull up the draft widget so you can send it but they removed the send button from the draft widget, presumably because they want to push you to send through the dot, which they decided can't send on your behalf
oh also, you can call the dot, it's just advanced voice mode packaged for ordinary people, but no thought was given as to making the voice match the appearance of the dot, or to making the voicemode connect quickly, so the experience leaves a lot to be desired; I don't even know what they get from making the voice mode connect slowly
it's a very strange product, it is very polished on the avatar, like incredibly polished, and then everywhere else, there's like no polish; maybe this is all about first mover advantage and they have really taken to heart that you have to be first, but first means the combination of application and model capability first, and the model has to be unhobbled, which this one is not
they really did nail the graphic design of dots, dots graphic design is really good, people are going to love it, or at least, their ordinary person personal assistant audience will love it, and that's really who its for, it's designed to make ai popular with people who don't use ai, it's not a dev tool really, even though as a ui/ux affordance, they also seem to want it to be a dev tool
but for it to really be a winner, it needs to be enterprise automation, because that's where the next big revenue spike is sitting, it's sitting in the hybrid of claude tag and computer use; so many enterprise tasks rely on this workflow, being able to nail it is a lot of tokens you can sell, you can get non-developers racking up claude code level bills for their companies
and you get the personal assistant; oh, and the personal assistant will need an additional layer on top and that is the ability to make phone calls on your behalf
2
10
956
the rationalist community was never going to be able to enter politics itself, as a whole, its whole strength is that it can develop out of the overton window ideas
4
74
ultrafast is the answer to all life's problems; the sand god injected into your blood at 300 tps;
1
58
the openai $200/mo plan just went from being a great deal to being ~equal to the anthropic plan in terms of value (or, maybe a little worse)
Hi,
Tomorrow we are re-opening the Pro $200 subscriptions to new subscribers, but together with it we are also changing how we calculate the usage for it. In effect, if you do the math, it will net out at half the dollar in API spend compared to the old Pro $200 plan.
Now that it's said, let me explain why this is happening and why you will still get more work done than if you were on the Pro $200 subscription one month ago.
(a) We didn't want to compromise in other ways and are committing to not reintroducing the 5h limit, so that you can fully use the weekly usage when you want.
(b) On the subscription, we guarantee that over time you always get more work done and with an increasing level of quality. This means that you will continue to get more value per dollar spent as a result of models getting more efficient and us passing down the improvements in the form of API price reductions.
(c) We don't want to put an incentive on ourselves to artificially inflate the API list prices to make it look like you are getting a lot (and workaround it through discounts, etc). Instead we want to continue to both rapidly reduce prices and increase capabilities of models on the API. This week we introduced GPT-6 Sol and GPT-6 Luna at 50% of their previous price. Over time, we see prices go low enough that it makes sense for most to buy usage as needed without there being a significant gap between what you get in a subscription and what you get in the API for a dollar spent.
(d) Tomorrow, we are adding more things to the subscription that won't draw on the usage, I won't reveal what that is yet.
I wanted to be transparent before all the big announcements tomorrow. Lots of new exciting things are coming to the subscriptions that will make it super compelling, but I wanted to make sure to share this change ahead of time so you can all understand it before we shower you with good news.
Codexingly,
Tibo
7
330
the beliefs of effective altruism are not that interesting; the interesting parts of it are its institutions and social dynamics;
1
70
ok i need to read more about this but i basically see this as against the LTBT; since my understanding is that the LTBT appoints all the board members now
but, they never revealed exactly how the LTBT could be undone, which always seemed a bit misaligned
ANTHROPIC SEEKS FOUNDER CONTROL AHEAD OF IPO
Anthropic is asking shareholders to approve a new structure that would give CEO Dario Amodei and the company’s six other co-founders a combined 50.1% of voting power on most corporate matters, per The Information.
The founders would hold the special voting shares through a separate LLC, allowing them to retain control despite relatively small economic stakes. Dario Amodei currently owns about 2%.
One major exception: Anthropic’s Long-Term Benefit Trust would continue to control the appointment of most board members, while the founders’ board seats would increase from 2 to 3.
Employees are also expected to receive a special class of stock that could serve as tie-breaker votes on certain issues.
Anthropic, valued at $965B in May, is reportedly targeting at least a $1.5T valuation in an IPO now expected around late October or November.
Source: The Information
3
368
If OpenAI releases a $500/month plan, I would consider it. But, it sort of feels like a hard sell because Anth and OpenAi trade the best model back and forth and right now Astra < Opus 5.5.
2
4
350
openai has never done a good release of a voice product, ever! they always announce it and then you have to wait a couple of days (or, worse) to get access
2
71
openai’s chatgpt app makes no sense on the chat side the models are 5.6 and 6 pro, whatever that is, and on the work side it’s 6 sol and astra but there is no live voicemode on the work side
the whole thing is a masterclass in useless complexity
2
124