THE AI BUBBLE HAS STARTED TO BURST
(twitter.com)
You're viewing a single comment thread. View all comments, or full comment thread.
Comments (35)
sorted by:
Claude has always been an overhyped, very high price-per-performance that everyone kept pushing based on raw benchmarks.
I keep telling people that OpenAI is far more economical than Claude because of this. I guess Microsoft just figured this out.
And the spin that this means "AI Bubble has burst" is so silly, even based on the headline in this screenshot itself!
It's not just Claude.
Now do you guys realise why datacenters and cheap energy are so crucial for the future?
More computational power with less energy cost = more token processing power.
He who can compute more and create the most powerful LLMs and solve the most powerful problems.
We are at that point in comparison to the PC evolution where we could barely fit 16KB of RAM on a PC. Today we can fit close to a TB of RAM on a PC. Thats a huge slop upwards and took 40 years to get to.
With LLMs the compute power will grow exponentially if Trump's plan works and will usher in centuries of prosperity and freedom.
Interesting then the direction Deepseek was going in.
BTW: I kind a like GAB as a meta AI. But indeed, Grok rules... on certain matters.
Elaborate?
From my perspective, it seems more that, with LLMs, the REQUIREMENT for computing power will grow exponentially, owing to the geometric growth of the network interconnections. Somebody needs to arrive at a cost-effectiveness metric for a process that may consume $ billions and produce intellectual morons.
I've seen some amazing hallucinations in AI-produced videos. I wouldn't want to have an AI-driven robot performing brain surgery on me. (Or MCAS flying an airplane, but that's another story.)
Hopefully this race to the top makes practical embedded NPUs viable. Ever been stuck at a red light for a while even though no cars are coming? Having an embedded NPU would enable you to not connect the camera to the cloud. Cheaper and more reliable infrastructure, more privacy, and less congestion to boot.
I highly predict that this is where things will head. Personal cloud.
I actually have a lot of my stuff including photos sync straight to my home pc via pangolin
Let's just see what happen here. I do like Grok.
Claude is actually pretty damn good. If people would stop acting like whiny liberals and actually learn to leverage these tools, we could all the Q said we should be doing at 1000x what we do now.
Its not whether Claude is pretty damn good - I told you that Claude consistently stays at the top of the benchmarks - but what the price for that little incremental extra performance (typically 3-6 months lead) ?
I started using Claude and burnt through the tokens very quickly. But then using GPT which is only slightly lower than Claude, cost me just $20 a month and did not affect the quality coding one bit - because I knew how to prompt it effectively, and its reasoning actually makes up for the base model quality.
So yes, Claude is a better model but too expensive, but GPT is by far the more cost effective model.
Its worse than being whiny liberals. Its this insane fear of technology because they learnt the wrong lessons from 2020.
Not sure what you meant to say there, but I find my productivity going up 50-100x easily.
We can spread memes 24/7 easily. That’s what I mean. Agree that Claude is over priced. I use it for QA in my agent swarm via my sub and the cli. Gpt 5.5 is my brain. Use locals for content gen
Exactly.
Microsoft worked out it's own ai is cheaper to use for it's purpose than Claude, and tbh I kinda shocked they didn't start with copilot and went with Claude in the first place.
The reason was that top many people inside were using Claude in their own so they decided to roll it out. The hype around Claude is insane. I have seen this first hand
Claude is better.
They had their engineers milk claude for all it's worth.
Now they will train Copilot on Claude's work.
Copilot will be as good and cheaper to run.