Hamza Adnan

Building WissenAI. Former LS Equity pod bp collector. Finance, Philosophy, AI. Some signal, mostly noise.

[Hamza Adnan @overfittedalpha]

The less assessable skills an MBA is supposed to confer (structured decision making, critical business thinking, incentive alignment) are genuinely useful and will continue to be even more important as other areas skills are commoditised by LLMs. But the degree is a noisy proxy and causality is blurred - recently main function of MBAs has been to correlate graduates with those skills, not cause them. Experience operating directly in business is a far better proxy, and increasing better RV in a “post-LLM” world.

[Hamza Adnan @overfittedalpha]

@sashabirukoff @latentleaps Thoughts on how long this took vs building something the customers asked for?

[Hamza Adnan @overfittedalpha]

@james_k_he @LacertaXG1 what part of the cycle is this?

[Hamza Adnan @overfittedalpha]

@LacertaXG1 AI Enterprise adoption finally picks up pace

[Hamza Adnan @overfittedalpha]

@latentleaps @aidenybai @samedotdev I’m feeling lucky

[Hamza Adnan @overfittedalpha]

@latentleaps And what does it mean

[Hamza Adnan @overfittedalpha]

@LacertaXG1 @latentleaps @HighyieldHarry Hmm would Sisyphus be happier pushing rock or slop uphill?

[Hamza Adnan @overfittedalpha]

@latentleaps @LacertaXG1 @HighyieldHarry When the VCs demand TAM expansion and you begin hallucinating use cases

[Hamza Adnan @overfittedalpha]

@latentleaps @AravSrinivas @LacertaXG1 I thought this was interesting, then I checked the numbers and sources - slop. Dangerous slop. Did no one at @perplexity_ai eval any of this before shipping it and claiming it’s a financial analyst? Move fast and deceive people

[Hamza Adnan @overfittedalpha]

@Complex Was Claude 4 piloting the plane

[Hamza Adnan @overfittedalpha]

If you’re an LP in Pabrai how can’t you be worried after seeing such high conviction intellectual laziness. Extrapolating a potential single factor overvaluation into an absolute, long-term outcome, ignoring the market’s historical ability to adapt and recover with certainty that approximates 100% is absurd.

[Hamza Adnan @overfittedalpha]

@_Angustina @NickLovesSpain You my good sir are a man of taste - best bottle under £10 in the UK

[Hamza Adnan @overfittedalpha]

@Nobu__11_ @hageverraz @METR_Evals You could equally extrapolate classifying trains to imply the improbable but it doesn’t change the fact that the y axis is stupidly defined

[Hamza Adnan @overfittedalpha]

@hageverraz @METR_Evals Something about extrapolating progression to an absurd conclusion…

[Hamza Adnan @overfittedalpha]

@cam_perrault Berkson’s paradox no? No positive characteristics will appear to be negatively correlated to you

[Hamza Adnan @overfittedalpha]

Great breakdown. Curious if you’ve tested how robust each model is under real world scenarios (eg handling poorly scanned PDFs, handwritten notes, or dense financial tables with subtle formatting variations). Is the performance gap between Mistral OCR and the premium or agentic modes amplified or narrowed in those contexts?

[Hamza Adnan @overfittedalpha]

Reminds me of Notion’s early trajectory - rapid adoption, steep valuation, scepticism around stickiness. Their success came from embedding deeply into workflows, making switching costs high. If Cursor manages workflow lock in (e.g. GitHub) then the valuation makes sense. Otherwise, it risks the Clubhouse effect where it’s viral today but commoditised tomorrow.

[Hamza Adnan @overfittedalpha]

@deedydas Having actually just tested the OCR API, I don’t know what benchmark targeting they’ve done but PDF parsing is still far from solved

[Hamza Adnan @overfittedalpha]

@unusual_whales Not as extreme but data out of other universities seems to support the trend. Unclear if changing pattern of employment demand, economic slowdown, first vintage competing against white collar AI products, or Bill Ackman…