> In other words, that usage you like is costing them tons of money
Evidence? I’m sure someone will argue, but I think it’s generally accepted that inference can be done profitably at this point. The cost for equivalent capability is also plummeting.
I didn't think there would need to be more evidence than the fact they are saying they need to spend $600 billion in 4 years on $13bn revenue currently, but here we are.
Right, but if OpenAI wanted to stop doing research and just monetize its current models, all indications are that it would be profitable. If not, various adjustments to pricing/ads/ etc could get it there. However, it has no reason to do this, and like all the other labs is going insanely into debt to develop more models. I'm not saying that it's necessarily going to work out, but they're far from the first company to prioritize growth over profitability
This meme needs to go in the bin. Loss making companies love inventing strange new accounting metrics, which is one reason public companies are forced to report in standardized ways.
There's no such thing as "profitable inference". A company is either profitable or it isn't.
Let's for a second assume all the labs somehow manage to form a secret OPEC-style cartel that agrees to slow training to a halt, and nobody notices or investigates. This is already hard to imagine with the amount of scrutiny they're under and given that China views this as a military priority. But let's pretend they manage it. These firms also have lots of other costs:
• Staffing and comp! That's huge!
• User subsidies to allow flat rate plans
• Support (including abuse control and handling the escalations from their support bots)
• Marketing
• Legal fees and data licensing
• Corporate/enterprise sales, which is expensive as hell even though it's often worth it
• Debt servicing (!!)
• Generating returns for investors
Inferencing margins have to cover all of those, even if progress stops tomorrow and the RoI to investors has to be likewise very large, so margins can't be trivial. Yet what these firms have said about their margins is very ambiguous. As they're arriving at this statement by excluding major cost components like training, it's not clear what they think the cost of inferencing actually is. Are they excluding other things too like hw depreciation and upgrades? Are they excluding the cost of the corporate sales/support infrastructure around the inferencing?
To be clear, it's absolutely impossible for OpenAI and the others to stop. The valuation and honestly the global markets depend on them staying leveraged to the hilt. So they're not going to stop. However, the point is that the models are genuinely useful and people pay for them, and if we reset the timeline with a company that has just the current proprietary models, they could turn a profit. That might involve charging more than they do now, etc. But this is much different than OpenAI, specifically, trying to turn a profit today, which wouldn't work for many reasons.
But also, "profitable inference" IS a thing! "Gross margin" is important and meaningful, even if a company has other obligations that mean it's overall not profitable.
"profitable on inference" means "marginal costs of inference are lower than revenue". It is very common to distinguish between upfront costs vs. marginal costs when judging the economic viability of a business.
You mention "debt servicing", but OpenAI has no debt. All the money they have raised is equity not debt.
Nope. The only "all indications" are that they say so. They may be making a profit on API usage, but even that is very suspect - compare against how much it actually costs to rent a rack of B200s from Microsoft. But for the millions of people using Codex/Claude Code/Copilot, the costs of $20-$30-$200 clearly don't compare to the actual cost of inference.
If this is like the flow it uses for a codex / ChatGPT subscription it doesn’t even register a handler - the redirect opens as a 404 in your browser and there are instructions in copying the token from the query string!
Second anecdote, I take between 10 and 15 grams. I don’t experience cognitive effects at lower doses (though my weightlifting endurance is still higher on lower doses). I also don’t eat meat so don’t have any incidental consumption
One of her creatine videos mentions that your muscles will take up ingested creatine faster than the brain. So for any creatine to make its way to the brain, your muscular creatine stores must be topped up first.
I think dosage would depend on the amount of daily physical activity. If you work out a lot, you'd have to replenish your muscular creatine stores before the brain could access any/much.
She also mentions boosting creatine dosage after bouts of mental exertion.
To add another data point, a 2024 study [1] on the mental effects of single doses of creatine was using 0.35g/kg of creatinemonohydrate, or about 28g for a typical adult male. Though obviously high doses are safer if you just do them once
And an earlier 2018 article [2] argued that "Evidence suggests that the
blood–brain barrier is an obstacle for circulating cre-
atine, which may require larger doses and/or longer
protocols to increase brain creatine as compared to
muscle. In fact, the broad spectrum of creatine sup
plementation studies that span different dosing pr-
tocols (e.g. high-dose short-term, low dose longer-
term), co-ingestion of other nutrients/compounds
(e.g. carbohydrate, protein, insulin), different popu
lations (e.g. vegetarians, elderly, patients, athletes)
is unavailable for brain creatine adaptations"
Meat is one of the primary sources of dietary creatine, but still provides overall very little (~2g/pound of uncooked red meat). There isn't much to make up for in a non-meat eater and the 5g should still be fine.
In testing for my workflows copilot significantly underperforms the SOTA agents, even when using the exact same models. It's not particularly close either.
This has lead to 2 classes of devs at my company a) AI hesitant, who for many copilot is their only interaction, having their worst fears confirmed about how bad AI is. b) AI enthusiasts who are irritated by dealing with management that don't know the difference pushing back on their asks for access to SOTA agents.
If I were the frontier labs, and wasn't billions of dollars beholden to Microsoft, I'd cut Copilot off. It poisons the well for adoption of their other systems. I don't deal with the other copilots besides the coding agent variants but I hear similar things about the business application variants.
Microsofts AI reputation is in the toilet right now, I'm not sure if its understood how bad it really is within the org.
Interesting - these head to head comparisons you’re doing with the same model - what harnesses are you comparing, say Claude code / codex versus copilot cli?
> I'm not sure if its understood how bad it really is within the org.
I can’t speak to that, but there’s a lively culture of people using internal tooling who also extensively use 3p products on projects outside work and are in a reasonable position to assess how well GH copilot works.
Yeah, I’m only interested in cli and non-interactive agent usage. I don’t compare say the vs code plugins, but do regularly compare say GitHub code reviews.
Those comparisons for instance have made us turn _off_ copilot pull requests entirely. All of the agents have false positives (as do humans) but copilot was having negative value in that context.
I’ve only started using it, so maybe I’m holding it wrong, but the other day I asked the IntelliJ plugin to explained two lines of code by referencing the line numbers. It printed & explained two entirely different lines in a different part of the file. I asked again. It picked two lines somewhere else.
After using ChatGPT for the last 6 months or so, Copilot feels like a significant downgrade. On the other hand, it did easily diagnose a build failure I was having, so it’s not useless, just not as helpful.
Sure I love Claude Code too - I use it plenty outside of work. But funnily enough I’ve been asking myself about whether to get my org on board with internal Claude Code trials and was struggling to truly articulate what we were losing versus the Copilot cli. There are some feature gaps - but the pace of work is super and experience is pretty good for me.
No one should hit Microsoft over the head for giving people access to Claude code - choice and competition is good!
I’d suggest doing some research on software quality. Two years back I was all for buying one (I was considering an EX40), but I got myself into some Facebook groups for owners and was shocked at the dreadful reports of quality of the software and it completely put me off. I got an ID4 instead. Reports about the EX90 have been dreadful. I was very interested, and I still admire their look and build when they drive by - but it killed my enthusiasm to buy one for a few years until they get it right.
Agreed that it can work well, but it can also irritating - I find myself using private conversations to attempt to isolate them, a straightforward per-chat toggle for memory use would be nice.
Love this idea. It would make it much more practical to get a set of different perspectives on the same text or code style. Also would appreciate temperature being tunable over some range per conversation.
Evidence? I’m sure someone will argue, but I think it’s generally accepted that inference can be done profitably at this point. The cost for equivalent capability is also plummeting.
reply