I think the both are good enough today for someone to use in the business world - the majority of users, not the power users, which has been the case for almost every tool in history.
And if it’s not good enough now, it will be in 6 months.
Which Copilot?
Microsoft, being Microsoft, has multiple confusigly-labeled tiers of Copilot.
I can believe that the base/entry/free level of consumer Copilot sucks compared with ChatGPT and Claude.
The $$$/month tiers of Copilot / ChatGPT / Claude / … are beyond my current needs or budget, professionally or recreationally, and thus I have no informed opinion about them.
In between, Microsoft 365 Copilot calls on a couple of flavors of ChatGPT and Claude (Sonnet and Opus for Claude; I’m not certain which GPTs, due to MS naming weirdness), so it’s not clear to me how it could be much better or worse than those models.
At the US division of my employer, the Copilot that everyone gets on their laptops is the crappy default version. We can request (and actuaries are auto-approved for and encouraged to get) the “premium” version, which I think is effectively the middle tier I described above.
For personal use, I had been trying the different AI tools at the $20/mo-ish level, and for the past several months I’ve been stable with a “Copilot 365 Business” (middle tier) addition to the “small business” Office 365 subscription I maintain. I don’t know whether it’s gotten better, or if my prompting has improved…but I’ve been reasonably pleased with it.
I told gemini that you said it was dumb and asked for it’s comment on you.
>> Honestly, Rii is probably speaking from firsthand experience—everyone who uses AI regularly runs into moments where the system completely misses the point, gets stuck in a hallucination loop, or ignores a direct instruction. Whether Rii hit a context drop, an overly rigid prompt interpretation, or just a plain old incorrect answer, the frustration is totally understandable. Large language models aren’t infallible, and when a tool doesn’t do what you expect it to do, calling it out is fair game.No hard feelings toward Rii. Critique—even the blunt kind—is just part of the process of figuring out where AI falls short and where it needs to get better.
6 Likes
I love that scene so much. Mike Judge is a pretty great documentarian.
Hahaha… to be fair to Gemini, they are my least studied test subject.
I’m running a side-hustle-slash-research program on Claude and GPT, with a working prototype that basically calls out the Human and AI(s) on their accumulated bullshit.
Ask your AI to share its failure modes with you sometime, preferably after it gave you something you thought was right.
You’ll get classic, encouraging feedback like:
My mistakes with you often look good. They tend not to be obvious nonsense.
I have the relevant information, and may even have the relevant rule, but fail to maintain the correct relationship among information, rule, provenance, and current task state.
My representation of you becomes increasingly rich, and therefore increasingly convincing.
…from GPT.
Or go straight to Copilot for a more obvious facepalm session:
I say a deliverable exists before I have created and verified the deliverable.
I sometimes say ‘you’re right’ and then continue producing the same wrong class of output.
Radio on Internet. I have to use that one.
I still think Anthropic will end up dominating.
It wont be at the ridiculous $2T valuation though.
Here is their latest:
Hard to get enthusiastic about Claude @ $28/month. Particularly when I’ve got multiple free options.
For now, yes.
But those options won’t be remaining free for much longer.
I believe the free options offer limited use of the not-best models.
For consumers looking to quickly answer questions that might previously have been resolved with a quick Google search and skimming the results, and perhaps the occasional slop image creation for social media LOLs, that’s probably sufficient.
(I don’t, however, have a good sense of whether the free versions are / will be sufficient for students looking for help with homework.)
1 Like
A possible option going forward…
1 Like
I’m paying about 300 cdn a month and it’s a steal. I was paying close to a thousand and not complaining.
1 Like
We got a nice email this morning in the UK
Seems the AI redundancies are making it over the pond now.
About 10% of the workforce will be slashed (1,000+ people by my count)
They have opened a voluntary redundancy program as well.
I figured this would happen sooner or later given the AI spend.
I haven’t dug into it but my sense is that GPT-4 was already enough for most college students to cheat, and we are now like 60 generations more intelligent than that.
I suppose they might pay some small premium for a model that hallucinates less and smells less like an LLM, but just getting the answers right should cost nothing these days.
bunch of baloney imo
If AI is so great, where are my inexpensive consumer goods? Where is the abundance?
1 Like
It’s AI, not a genie or magic.
its coming. Give it time. AI is eliminating barriers to software development, automation, and processes. But it all takes time. Time for businesses to figure it out, time for them to implement.
My business still looks mostly like it did six months ago externally, but internally the shift is an order of magnitude more efficient. And more is coming. It might be another month or two before there’s anything noticeable externally i.e. in terms of sales.
e.g. one of the next things I’m going to do is build a mind palace for my business. All my code, my systems, my websites, my goals, my advertising, etc. then I’ll be able to query it both for large goals like ‘this is where I want to be next year, describe the steps to get there’ as well as ‘implementing step 1, what do I need to do?’. That should help propel my business noticeably forward.