They Just Admitted The AI Bubble Is Popping...
Credibility score: 46/100 — Mixed Credibility. Several questionable claims detected. Watch with healthy skepticism.
BSmeter analyzed "They Just Admitted The AI Bubble Is Popping..." and rated it 46/100 for credibility (a BS score of 54/100 — mixed credibility), on 2026-06-05. Its weakest claim — "Anthropic released a report today claiming AI reached recursive self-improvement" — scored 25/100 and was flagged as sketchy. 17 claims were checked against the video transcript. Scores are produced by BSmeter's AI analysis of the transcript, not independent human verification.
Of 17 claims analyzed: 3 scored under 40, 13 between 40 and 69, and 1 at 70 or above.
Claims analyzed
Uses local AI models on his own hardware — Personal Story (50/100)
At 0:30
Straightforward personal workflow claim.
Why this score: Personal claim about his setup.
- No external evidence needed
- Matches common local AI usage patterns
- Web context shows local models are freely available
Original quote: “Now, to give you an idea, I use artificial intelligence when I make these videos in terms of showing you how the technology works. Right now, you can see that the AI that I'm using is loaded onto my own local system. Yes, I do download local models, freely downloadable by you, me, and anybody else.”
This is simply how AI token pricing functions — Opinion (50/100)
At 2:30
Just Vibes
Why this score: - Descriptive framing, not a testable claim
Original quote: “That's just how it works.”
Manager claims $1M monthly AI spend at 5-6k employee company — Personal Story (50/100)
At 4:30
Unverifiable Reddit anecdote — no company named.
Why this score: Unverifiable personal claim.
- No company, date, or receipt provided
- Reddit posts are easy to fabricate or exaggerate
- Cannot be checked against public records
Original quote: “developers, these people have basically been saying, I'm a manager at a company with like 5 to 6,000 employees, and basically over the last couple of months, their AI spending has actually exploded. This was posted 5 days ago. So to give you an idea, the last month bill at this person's company…”
Uber imposed $1,500 monthly AI cap per employee — Dubious (40/100)
At 4:52
No public record of Uber announcing $1,500 AI caps.
Why this score: No supporting evidence found.
- Uber has not publicly disclosed per-employee AI spending caps
- Search results show no confirmation of this specific policy
- Claim relies on speaker's "I've talked to developers" without sources
Original quote: “Uber ended up actually having to issue caps on all of the AI that they would use inside their company because people were reaching its max. So, for instance, Uber puts a $1,500 cap per actual employee and the tools they use to write code, whether that be clot code or cursor.”
Uber burned through entire yearly AI budget in 4 months — Unverifiable (50/100)
At 7:06
No source named for this Uber number — just dropped it.
Why this score: Unverifiable claim.
- No report, internal document, or Uber statement referenced.
- Community comments treat similar assertions as recurring unproven narrative.
- Without a verifiable source, score defaults to neutral.
Original quote: “they had a budget for one year for people to use AI slop in their company. And it turns out in a matter of four [__] months, four months into the year, they actually burned through the entire [__] BUDGET.”
Microsoft finds AI coding tools cost more than human salaries — Dubious (40/100)
At 7:19
Paywalled report mentioned but never shown or named — convenient.
Why this score: Weak sourcing.
- Speaker admits it's summarized from a paywalled report.
- No specific Microsoft statement, earnings call, or document cited.
- Recent coverage (June 2026) discusses high AI costs but contains no confirmation of this exact comparison.
Original quote: “Even companies like Microsoft... also says that using this technology is actually more expensive than paying real [__] human people their salaries. Why? Because the prices have just gone through the roof.”
Microsoft moved thousands of devs from Claude to GitHub Copilot due to cost — Unverifiable (45/100)
At 7:35
Specific internal move with zero receipts provided.
Why this score: No supporting evidence.
- No Microsoft announcement, internal memo, or news report confirms this mass migration.
- "Quad" (Claude) usage details are paywalled per speaker.
- Web context shows ongoing cost debates but nothing matching this timeline or scale.
Original quote: “they've been moving people away from again the quad guys over to their own GitHub guys. And even though Quad was only open six months ago, thousands of their developers... were basically moved over because ultimately... it is actually too expensive”
AI causes net losses via hallucinations and hidden compliance costs — Dubious (40/100)
At 8:30
Real issue exists but framed as universal failure.
Why this score: Hallucinations and review overhead are documented problems.
- Studies show AI can increase time spent fixing outputs.
- Speaker generalizes this to 'a lot of these companies' losing money with no specific examples or data.
Original quote: “You see where the real problem with a lot of this AI slop is is that when a lot of these companies move their people over, they actually end up losing a lot of money, not because the AI is necessarily terrible. It's because they're dealing with hallucinations. They're dealing with the actual hidden…”
Claude prices shown on screen are the current rates — Solid (75/100)
At 10:30
✅
Why this score: - Speaker directly references on-screen pricing display
- No contradiction with known Anthropic pricing structure
Original quote: “if you look at the prices for any of these models, for instance, if you go to like Claude, you can actually see that their prices are kind of a as we've seen right over here”
Claude top-tier models cost $25 per million tokens — Dubious (45/100)
At 10:36
Aggressively rounded — current top rate is $15/$75 split
Why this score: $25 flat rate is incorrect.
- Anthropic's latest Claude 3.5/Opus pricing is $15 per million input, $75 output
- No current tier charges a flat $25/million across the board
- Speaker conflates blended or hypothetical cost with listed rate
Original quote: “they've given you their prices based on input tokens, cash rights, cash hits, output tokens, yada yada yada. Now, this is by far one of the most like expensive AIs to kind of play around with because again, if you're looking at things like their most, you know, top tier models, you're looking at…”
Deepseek offers AI access for pennies vs US models — OK (60/100)
At 10:48
True directionally, numbers are vibes
Why this score: - Deepseek pricing is substantially lower than US frontier models
- Exact "pennies" figure never specified — remains impressionistic
- No receipts given for precise cost comparison
Original quote: “Deepseek as a company has basically provided access to their artificial intelligence for, I'm not joking with you, pennies compared to, again, their big American counterparts”
AI IPOs will be a rug pull leaving new investors holding the bag — Sketchy (35/100)
At 12:30
Rug-pull framing is dramatic — IPOs aren't exit scams by definition 💀
Why this score: Rug pull implies deliberate fraud and sudden collapse.
- IPOs involve regulatory scrutiny and disclosures.
- Losses from overvaluation are possible but not the same as crypto exit scams.
Original quote: “And so in crypto terms, you know, I I think this is what we kind of call like a weird ruck pull where I I think a lot of these companies, you know, when the initial investors pull their money out and the new guys who put their money in, you know, the exit liquidity as some would call it, that's…”
Anthropic released a report today claiming AI reached recursive self-improvement — Sketchy (25/100)
At 14:30
No such Anthropic report exists today — pure hype timing for the IPO 💀
Why this score: No matching report. Key issues:
- No Anthropic release on recursive self-improvement appears in today's coverage
- The speaker ties it directly to IPO hype without evidence
- Real-time search shows no confirmation of the claimed announcement
Original quote: “Anthropic released this like report today and said when the AI builds itself, so we're basically reaching what we call recursive self-improvement.”
AI companies compare their tech to nukes; Pakistan/India built bombs via smuggling — Dubious (45/100)
At 16:30
Nuke comparison is real — smuggling claim is sloppy history.
Why this score: Nuke analogy exists in AI safety discussions. Problems:
- Pakistan's program involved A.Q. Khan's theft from Europe, not simple 'smuggling'
- India's program was largely indigenous with some dual-use imports
- Speaker hedges with 'I believe' and 'in certain cases' while stating it as fact
Original quote: “they do compare their tools to something like missiles and nuclear weapons... Pakistan and India... developed their stuff, I believe, in certain cases by smuggling the parts”
AI bubble exists because companies fake demand with user counts while hiding real costs — Dubious (35/100)
At 18:30
User metrics ≠ fake demand. Costs are public.
Why this score: Conflates engagement metrics with fabricated demand.
- Active users are real usage, not invented numbers
- Major labs publish quarterly infrastructure spend (hundreds of millions to billions)
- JPMorgan analysis (Dec 2025) found AI investments don't meet classic bubble criteria
Original quote: “This bubble is only possible because these people are able to get billions of people to use their AI models. Even if you've ever just interacted, you count as a number. And that number is used to show a lot of these investors that the demand exists. But these investors don't really understand, I…”
Says AI bubble is popping because current models aren't efficient — Opinion (50/100)
At 20:30
Calling it a 'pop' while the market's still running hot 💀
Why this score: Opinion on market trajectory.
- Speaker frames inefficiency as proof of imminent collapse
- No specific timeline or threshold given
- Web results show ongoing debate — JPMorgan and others still reject classic bubble criteria as of late 2025
Original quote: “it's just not efficient. And because it's not efficient, that is where the real actual pop sort of comes. That's why I'm telling you the bubble pop is a thing.”
Says the AI bubble is close to popping — Opinion (50/100)
At 22:30
Everyone's been calling the pop for months — still waiting on the receipts 💀
Why this score: Classic bubble narrative with no new evidence.
- Web results show ongoing debate; JPMorgan says AI doesn't meet classic bubble criteria
- December 2025 analysis found investments tied to real productivity gains, not pure speculation
- Community comments echo the fatigue: "AI bubble popping for the 27th time this week"
Original quote: “it's getting too expensive. The bubble is actually [__] close to popping at this point.”
See the full analysis with timestamps →