Anthropic Brings Automatic Memory to Claude Pro and Max Users

It also doesn't remember when I tell it not to use – in a sentence and I even put it in its writing style rules on the Profile settings.

a persistent and accurate memory seems to be incredibly difficult for LLMs

In future they probably will never forget anything and that is quite creepy. Like a wife that never forgets that you forgot her birthday 17 years ago.

I think a better route than going under or being acquired would be if they pivot and just work n research. Just focusing on interpretability as perhaps a non-profit would be cool. I love reading about their work in that sector.

But yeah, they won't make it past 2026.

Feel free to @ me on Jan 1 2027. I'll happily eat crow, and I don't think you were replying based on factual information which is a little disingenuous. Nothing personal, seriously, but... having a 'feeling' about it and using data points which don't logically follow isn't a good debate tactic, and it especially isn't good analysis.

If you're talking more like 2030 timeframe, possibly some of these companies will fail, but I think it's going to be the myriad wrapper companies (e.g. every AI 'company' on linkedin), not the foundation model companies – and particularly the ones who are publishing research regularly – that are going to go under.

We absolutely are going to hear about an "AI crash" when those companies fail, and "AI winter" and all of that nonsense, but the foundation companies who have a solid backing and more importantly solid research are going to soldier on. I think Anthropic was in a much tighter spot a year ago vs. now, and they've been firing on all cylinders for the past year with measurable, regular improvements.

The people on reddit and elsewhere who complain are using the free or in rare cases the $15/mo versions and are upset they can't get the crazy usage they can out of a (very bad) version of ChatGPT.

...

On-topic for the thread, I actually don't like this memory feature very much for certain use cases and think it is too tuned for extreme business cases, I had to modify my core project's memory ~5+ times yesterday to get it to have something in there that was coherent to what I'm working on, but for Joe Coder it will be very useful.

If they offer doubled context with the Max plans, even at $200/mo, I'd upgrade immediately from my current Max.

So, anyone using this – make sure you actually check what's in the memory. I HOPE they let us disable memory per project eventually because it's much more useful for business right now and kind of harms other things where you can use project files to do accomplish more at a much more directed and granular level.

Do you really think the bonfire of cash will last forever and no one will ever come to their senses? Meta just laid off 600 of their own AI employees and would continue to further fund a company with no future other than burning cash?

Incredibly doubtful they make it past 2026 without being acquired. Let's put a pin on this and revisit this then.

You're putting a pin in the discussion after quoting / making up a bunch of stuff with absolutely no basis in fact.

The lawsuit was more than covered by a funding round and gives them the legal ability to use written text as training data in a way they can be forward about. Right now you need to kind of work around a lot of limitations to get it to tell you information, but it is in there. Recouping that cost will probably take a while but they have significantly less legal exposure now vs. OpenAI who still has ongoing legal action. But they won the case, and they have the technical infrastructure already implemented to scan a very large corpus vs. just scraping reddit and forums for nonsense.

As MLVC else said they also have the backing of Amazon, and a founder who was stubborn about selling and instead forged partnerships for financing vs. a buyout. Sonnet 4.5 is incredibly good and Opus 4.1 is unmatched for many things, and I speak from direct experience having spent a lot of time using all the major models except Gemini.

I don't think you use Claude to any kind of serious degree and are parroting pundits who don't know what they're talking about, which is nearly everyone in the "AI commentary" space. Speculation is fine but you might want to look more at the details, as well as their research – and their paid user / inference profitability.

They are making profits on the Max plan users and have throttled Pro down so far they likely are there too. ChatGPT has more users but I speculate there are actually more paid users of Claude vs. ChatGPT (which is sub-5%, and is why they are pivoting to ads and adult content).

I know many coders who happily pay for Max plans or direct credits. Their R&D is also advancing the field dramatically, and the blog posts about things are not hyperbole despite the average anti-AI bandwagon internet person thinking it is just marketing material. I've seen what they're talking about first hand and it's groundbreaking.

I think your usage stats are absolutely bogus, especially since non-paid users of ChatGPT are not generating any revenue at all right now (which will change of course). When asked to source them you didn't. Considering doing so, I'm specifically interested if you have information that "it has about the lowest traffic in AI", particularly when it comes to paying users.

I will absolutely take whatever bet you're implying. :)

It loses big money on everything it does, it has about the lowest traffic in AI, it already has a huge judgement against it in hustling one of their many lawsuits. And unlike all the others it doesn't have a Microsoft, Facebook, Google or Musk with all the money if the world burn on in a giant bonfire on it. Surely someone will acquire it, if not acquihire it, but I thin it has too much baggage in Dario and their his co-founder.

I serious doubt they make it through 2026 and not be acquired. And since there is no moat, especially with every latest AI methodologiy aging like milk, and with the employee loyalty of floozies, why buy the cow?

Where are you getting this information? Anthropic has Amazon as its largest investor with $8 billion committed, plus Google owns roughly 10-14% of the company with over $3 billion invested. They just closed a $13 billion funding round in September at a $183 billion valuation, and their revenue run rate is approaching $7 billion annually.

I'm genuinely curious where you're seeing information about them not making it through 2026, because everything publicly available shows the exact opposite. Can you share your sources?

will it remember that like the vision pro only nerds care and the product is overpuffed?