A browser doesn't count as a significant business integration?
It seems to me the author's meaning of business integration is pretty clear: "profitable workplace integrations, such as a serious integration with Microsoft office suite or the google workplace suite." -- integrations tailored to a particular company's offerings. A browser is not business-specific nor workplace-specific.
I don't usually venture into LessWrong culture or "feels" issues, but the above comment seems out of place to me. As written, the comment seems more like something I would see on Hacker News, where someone fires off a quick quip or a zinger. I prefer to see more explanation here on LW. I'm in some ways a HN refugee, having seen too many low-effort ping-pong sniping contests.
For what it's worth, I didn't interpret the comment as a "zinger". While I thought I explained what I meant by "significant business integration", at least 2 other people upvoted the original comment, so I can only assume I didn't explain my point maximally well.
I'm happy the comment was made, and think that sometimes a short question can get to the point better than a long one.
It's non-zero business integration, but I'm thinking about something closer to fine-tuning on manipulating excel spreadsheets, collating information from Teams/Slack/Outlook/etc, creating power point slides from existing sources in the company. If you compare the AI-powered tools for SW engineers vs the tools for AI-powered non-SW white-collar workers, the difference is night and day IMO. Many of the existing business integrations are little better than copy-pasting the context into your favourite LLM, but (for example) cursor is significantly better than that, has custom-trained models for Tab autocomplete, etc.
Note: I'm writing every day in November, see my blog for disclaimers.
There’s been some questions raised about why AI hasn’t seen more widespread adoption or impact, given how much better ChatGPT et al. are above the previous state of the art for non-human cognition. There’s certainly been lots of progress, but the current era of AI is a few years old at this point and many low-hanging fruit which feels doable with current technology is not in fact, done.
Given this is what we see, I’m fairly confident that the frontier AI companies are intentionally not pursuing the profitable workplace integrations, such as a serious[1] integration with Microsoft office suite or the google workplace suite.
The reason for this, is if you had a sufficiently good AI software developer, you could get a small army of them to write all the profitable integrations for you ~overnight. So if you’re looking at where to put your data centre resources or what positions to hire for or how to restructure your teams to take advantage of AI, you emphatically do not tell everyone to spend 6 months integrating a 6-month-out-of-date AI chatbot into all of your existing products. You absolutely do pour resources into automating software engineering, and tell your AI researchers to focus on programming ability in HTML/CSS/JS and in Python. This, not coincidentally I’d argue, is what we see: most of the benchmarks are in Python or some web stack. There is also a significant amount of mathematics/logic in the benchmarks, but these have been shown to improve programming ability.
So what would we predict, if the above is true? I predict that ~none of the labs (Anthropic, Google, Facebook, OpenAI+Microsoft) will launch significant integrations that are designed for immediate business use-cases until either most of the code is written by their internal LLMs or if they see these products as a useful means of collecting data (e.g. Sora).
If this is true, then we can also infer that the leadership and stakeholders of these companies (if it wasn’t already obvious), is very AGI-pilled, and whoever’s pulling the shots absolutely believes they’ll be able to build a synthetic human-level programmer within five years. It doesn’t say anything about ASI or the AI’s ability to perform non-programming tasks, so it’ll be interesting to see if the movements of these big companies indicates that they’re going for ASI or if they just see the profit of automating software development.
While automating AI R&D is an explicit goal of some of these companies, I’m not sure whether this goal will survive the creation of a competent, cheap, human-replacement software developer. Up until this point, the steps towards “automating AI R&D” and “automating software development” are approximately the same: get better reasoning and get better at writing code, using software development tools, etc. But I’d argue that AI R&D is significantly harder than profitable software development. So for now, the companies can use the sexy “we’re automating AI R&D” tag line, but once a company builds a synthetic software developer I’m fairly certain that the profit-maximising forces at be will redirect significant resources towards exploiting this new-found power.
There do exist MS office/google workspace integrations, but they're about as minimal as you can get away with and still the pointy-haired managers "yes MS office 'has AI'". These integrations are not serious, they are missing a lot of very basic functionality that leads them to be little better than copy-pasting the context into your chatbot of choice.