Saying LLMs have only incrementally improved is like saying my 13 year old has only incrementally approved over the last 5 years. Sure, it's been a set of continuous improvements, but that has taken it from a toy to genuinely insanely useful.
Personally, deep research and o3 have been transformative, taking LLMs from something I have never used to something that I am using daily.
Even if the progress ends up plateauing (which I do not believe will happen in the near term), behaviors are changing; OpenAI is capturing users, and taking them from companies like Google. Google may be able to fight back and win - Gemini 2.5 Pro is great - but any company sitting this out risks being unable to capture users back from Open AI at a later date.
> any company sitting this out risks being unable to capture users back from Open AI at a later date.
Why? I paid for Claude for a while, but with Deepseek, Gemini and the free hits on Mistral, ChatGPT, Claude and Perplexity I'm not sure why I would now. This is anecdotal of course, but I'm very rarely unique in my behaviour. I think the best the subscription companies can hope for is that their subscribers don't realize that Deepseek and Gemini can basically do all you need for free.
I doubt it. Google is shoving Gemini on everyone’s face through search, and Meta AI is embedded in every Meta product. Heck, instagram created a bot marketplace.
They might not “know” the brand as well as ChatGPT, but the average consumer has definitely been exposed to those at the very least.
DeepSeek also made a lot of noise, to the point that, anecdotally, I’ve seen a lot of people outside of tech using it.
I can't square how OpenAi can capture users and presumably retain them when the incumbents have been capturing users for multiple decades and why can they not retain them?
If every major player had an AI option, i'm just not understanding how because OpenAi moved first or got big first, the hugely massively successful companies that did the same thing for multiple decades don't have the same advantage?
Who knows how this will play out, but user behavior is always somewhat sticky and OpenAI now has 400M+ weekly active users. Currently, I'm not sure there is much of a moat, as many would jump if, say, Google released a model that is 10x better. However, there are myriad ways that OpenAI could slowly try to make their userbase even stickier:
1. OpenAI is apparently in the process of building a social network.
2. OpenAI is apparently working with Jonny Ive on some sort of hardware.
3. OpenAI is increasingly working on "memory" as a LLM feature. Users may be less likely to switch as an LLM increasingly feels like a person that knows you, understands you, has a history with you, etc.
4. Google and MSFT are leveraging their existing strengths. Perhaps you will stick with Gemini given deep integration with Android, Google Drive, Sheets, Docs, etc.
5. LLMs, as depressing as this sounds, will increasingly be used for romantic/friend purposes. These users may not want to switch, as it would be like breaking up and finding a new partner.
6. Your chat history, if it can't be easily exported/imported, may be a sticky feature, especially if it can be improved (e.g. easily search, cross-reference, chats, like a supercharged interconnecting note app with brains).
I could list 100 more of these. Perhaps none of the above will happen, but again, they have 400M weekly users and they will find ways to keep them. It's a lot easier to keep users that have a habit of showing up, then getting them in the first place. There's a reason that Google is treating this like an emergency; they are at serious risk of having their search cash cow permanently disrupted if they don't act fast to win.
Very thought provoking reply. #3 sounds the most sticky to me, in the product sense that you'd build "your own LLM/agent" and plug it other services. I heard this on a product podcast [1], think of it like Okta SSO integration: access controls for your personal/sensitive LLM stuff vs all other services trying to get you to use their LLM.
#5 stands out as well as a substantial barrier.
The rest to me our sticky, but no more uniquely sticky than any other service that retains data. Like the switching cost of email or a browser. It does stick but not insurmountable and once the switch is made, it's like why did I wait so long? (I'm a Safari user!)
6 (can’t export/import chat history) is already a wrap since every user is prohibited from using ChatGPT chat logs to “develop models that compete with OpenAI,” if you export your chats and give it to Gemini or Claude or post it on X and Grok reads it, then you just violated the OpenAI terms, that’s grounds for a permaban or lawsuit for breach of contract (lol) … maybe your companies accept this risk but I’m in malicious compliance mode
Google is alright, but they have similar stupid noncompete vendor lock in rule, and no way to opt out of training, so there’s no real reason to trust Google. Yeah they could ship tool use in reasoning to catch up to o3, but it’ll just be catching up and not passing unless they fix the stupid legal terms.
Claude IDK how to trust, they train on feedback and everything is feedback, and they have the noncompete rule written even more broadly, dumb to use that.
Grok has a noncompete rule but also has a way to opt out of training, so it’s on the same tier of ClosedAI. I use it sometimes for jokey toy image generation crap but there’s no way to use it for anything serious since it has a copypasted closed ai prohibition
Mistral needs better models and simpler legalese, it’s so complicated and impossible to know which of the million legal contracts applies
IMHO meta is the only player, but they shot themselves in the foot by making Llama 4 too big for the local llama community to even use, super dumb, killed their most valuable thing which was the community.
That means the best models we can use for work without needing to worry about a lawsuit, are Qwen, and DeepSeek distills, no American AI is even in the same ballpark, Gemma 3 is refusal king if you even hint at something controversial. basically, America is getting actively stomped by China in AI right now, because their stuff is open and interoperable, and ours is closed and has legal noncompete bullshit, what can we actually build that doesn’t compete with these companies? Nothing
No, it's still just a toy. Until they can make the models actually consistently good at things, they aren't going to be useful. Right now they still BS you far too much to trust them, and because you have to double check their work every time they are worse than no tool at all.
It's been five years. There is no AI killer app. Agentic coding is still hot garbage. Normal people don't want to use AI tools despite them being shoved into every SaaS under the sun. LLMs are most famous among non-tech users for telling you to put glue into pizza. No one has been able to scale their chatbots into something profitable, and no one can put a date on when they'll be profitable.
Why are you still pretending anything is going to come out of this?
To extend your illustration, 5 years ago no one could train an LLM with the capabilities of a 13 year old human; now many companies can both train LLMs and integrate them into products.
> taken it from a toy to genuinely insanely useful.
Personally, deep research and o3 have been transformative, taking LLMs from something I have never used to something that I am using daily.
Even if the progress ends up plateauing (which I do not believe will happen in the near term), behaviors are changing; OpenAI is capturing users, and taking them from companies like Google. Google may be able to fight back and win - Gemini 2.5 Pro is great - but any company sitting this out risks being unable to capture users back from Open AI at a later date.