Microsoft JUST BROKE OpenAI...
Credibility score: 49/100 — Mixed Credibility. Several questionable claims detected. Watch with healthy skepticism.
BSmeter analyzed "Microsoft JUST BROKE OpenAI..." and rated it 49/100 for credibility (a BS score of 51/100 — mixed credibility), on 2026-06-04. Its weakest claim — "Fine-tuned Microsoft models will be 10x cheaper than GPT-5 with proprietary moat" — scored 35/100 and was flagged as sketchy. 19 claims were checked against the video transcript. Scores are produced by BSmeter's AI analysis of the transcript, not independent human verification.
Claims analyzed
Microsoft is swinging hard with new models and an OpenClaw partnership; Bernie wants to tax AI IPOs. — Just Vibes (50/100)
They're throwing everything at the wall this week! — Microsoft breaking up with OpenAI *and* Bernie Sanders wanting to tax them? Wild combo. 🤯🤖💰
Scorsese using Flux AI to produce actual Hollywood films — Dubious (40/100)
Storyboard tool ≠ film production — wording doing heavy lifting 💀
Big hedge funds sold Microsoft stock over AI fears — Dubious (45/100)
Vague 'a lot of big hedge funds' — zero names or filings cited 💀
Microsoft's strategy of trailing the frontier is potentially killer — Opinion (50/100)
Bold prediction — we'll see if the 'killer' part lands 💀
Microsoft's inference chip announced January 2026 — OK (60/100)
Timing matches recent reports but details still thin.
Microsoft introduced Frontier Tuning for custom models — Dubious (45/100)
Sounds like rebranded fine-tuning with zero public proof.
Microsoft's model benchmarks are self-reported only — Opinion (50/100)
Fair skepticism — third-party testing hasn't happened yet.
Anthropic faced major backlash over pirated training data — OK (65/100)
True-ish but timing and scale feel overstated.
Enterprises can use general 'frontier models' that are universally smart. — Solid (75/100)
They're all just... good? — That sounds like marketing fluff, but the premise is sound. 🤔👌
Custom models reach near-SOTA performance with 10x efficiency — Dubious (45/100)
Big claim, zero numbers or benchmarks shown — just vibes so far ⚠️
Fine-tuned Microsoft models will be 10x cheaper than GPT-5 with proprietary moat — Sketchy (35/100)
10x cheaper sounds great — until you realize no pricing or model actually exists yet 💀
Custom frontier tuning will outperform off-the-shelf models for enterprises — Opinion (50/100)
Classic "trust the process" take — heavy on faith, light on proof 🤷♂️
Microsoft has a new chip called Maya 200 — Dubious (35/100)
No Maya 200 chip — sounds like a mix-up with existing hardware 💀
Microsoft's Frontier Tuning lets companies train custom models on their private data — OK (55/100)
Sounds plausible but still marketing copy — actual results TBD
Microsoft's new model fixes enterprise control, cost, and data issues better than closed or open-source options — Opinion (50/100)
Classic vendor-lock-in pitch dressed up as enterprise salvation 💼🔒
Maya silicon makes AI training cheaper — Dubious (45/100)
No public pricing or benchmarks shown yet 💸
Microsoft creating MXC sandbox containers for AI agents in Windows — Dubious (45/100)
No public evidence of 'MXC' containers yet 🚩
Defines autopilot AI as fully autonomous 24/7 agents — Dubious (45/100)
Calling background agents "24/7 autonomous" while current tools still need heavy human oversight.
Microsoft internal doc calls phase one 'make people addicted' — Unverifiable (50/100)
No public copy of the document — just the speaker's word.
See the full analysis with sources and timestamps →