codebyaditya

há 6 mesesdiscuss

Hey, I didn't generate everything by AI but only to frame the words since i was busy with something else. I'll make sure from now on such mistakes will never happen again.

sincerely.

codebyaditya

há 6 mesesdiscuss

Hey, lemme apologize first. I indeed generated this above comment by AI because I was extremely busy with something else and didn't have time to respond but i'll make sure this will not happen again.

sincerely.

codebyaditya

há 6 mesesdiscuss

Excellent point. My analysis calculated Level I concentration within each employer type but not absolute volumes.

You're correct that if staffing firms file significantly more total applications, their absolute number of affected Level I positions could exceed product companies despite lower concentration.

To properly claim "3x harder hit," I'd need to show either: 1. Total Level I applications by employer type, or 2. Total estimated selection losses by employer type

Without those absolute numbers, the "3x harder" claim overstates what the data shows. The accurate claim is that product companies have 3x higher Level I concentration - but that's not the same as 3x more impact.

This is a good catch. The proportions tell us about hiring patterns but not total system impact.

codebyaditya

há 6 mesesdiscuss

Author here.

Quick summary of probability shifts under the weighted system:

  Level I ($80K median):  -57% selection probability
  Level II ($103K median): -14%
  Level III ($135K median): +29%
  Level IV ($158K median): +73%

The counterintuitive finding: product companies (direct employers) have 22% Level I concentration vs 7% for staffing firms. The rule designed to stop "lottery abuse" hits direct employers 3x harder.

Data source: DOL LCA disclosure data. Happy to discuss methodology.

codebyaditya

há 6 mesesdiscuss

(Methodology): Author here. Technical methodology: The counterintuitive finding: DHS justified the weighted lottery by citing "employers submitting large volumes of low-wage registrations"—understood to mean staffing firms. But our data shows:

Product companies (direct employers): 22.0% Level I concentration Staffing/consulting firms: 7.1% Level I concentration Ratio: 3.1× Tech companies hiring junior engineers get hit 3x harder than the outsourcing firms the rule supposedly targets.

Data: DOL H-1B LCA Disclosure Data FY2024, 517,874 certified applications. Stack: Python, pandas 2.1.0.

codebyaditya

há 6 mesesdiscuss

Fair point. To be clear we aren't saying wage should be the only rule, just that it's weird to see such a huge disconnect. Usually money talks.

Your "gateway" hunch is likely spot on. Most rural docs start on J-1 waivers (mandatory 3 years). If they actually stayed after that, we'd see way more H-1B conversions filed by those rural hospitals to keep them. Since the volume is so low, it suggests once the 3 years are up, they bail for the city. The wage premium just isn't enough to anchor them.

codebyaditya

há 6 mesesdiscuss

Author here. Quick methodology notes:

Data: 20,225 H-1B LCA disclosures from DOL, FY2024, healthcare occupations only Analysis: Python (pandas), mapped ZIP → RUCC codes, median wage by volume quintile Key limitation: This is LCA data (intent to hire), not final USCIS approvals

Interesting rabbit holes:

Urban/rural split isn't binary—codes 4-6 show gradient effects

Wage level inversions strongest in codes 7-9 (most rural)

Happy to answer methodology questions.

codebyaditya

há 6 mesesdiscuss

We analyzed the FY2024 H-1B LCA dataset to look for geographic inefficiencies.

Methodology: We joined DOL disclosure data with USDA Rural-Urban Continuum Codes (RUCC 2013) using a ZIP-to-County crosswalk.

The Signal: We found a clear inversion of the standard supply curve. Rural areas offered a 21.4% wage premium ($250k median vs $206k) yet achieved 10.2x lower placement volume.

Systemic Friction: The data suggests that for high-skill labor (physicians), geographic friction and regulatory overhead (including the new $100k fee) outweigh significant monetary incentives. The market is not clearing.

codebyaditya

há 6 mesesdiscuss

What I appreciated here is how calmly Tao separates useful pattern matching from actual mathematical understanding. There’s no AI hype or dismissal but just a reminder that proof, verification, and intuition are different things. It made me rethink where LLMs genuinely help vs where they just feel convincing. Thank you for sharing!

codebyaditya

há 6 mesesdiscuss

The disturbing part isn’t that bad encounters happen — it’s that these techniques are officially banned, yet keep showing up across unrelated cases. When policy and on-the-ground behavior diverge this consistently, it stops looking like individual misconduct and starts looking like a systems problem.

codebyaditya

há 6 mesesdiscuss

I read it less as obliviousness and more as internal language leaking into marketing. What’s “Liquid Glass” to Apple reads like an aesthetic system though but to outsiders it sounds like jargon inflation. I feel the gap between internal coherence and external clarity shows up in these releases a lot.

codebyaditya

há 6 mesesdiscuss

You’re right on the numbers....Firefox never had majority share. The stronger claim is causal influence, not dominance. I recently read somewhere that the Firefox (and later Chrome) forced standards compliance and broke IE’s de-facto monopoly mindset. IE’s decline was gradual and multi-factor, but Firefox clearly shifted developer and user expectations.

codebyaditya

há 6 mesesdiscuss

What’s unsettling here isn’t any single policy, but the convergence: predictive policing, protest restrictions, and administrative punishments all justified as “risk management.” Even if each tool seems narrow, together they normalize acting on suspicion rather than action, which quietly lowers the bar for dissent.

codebyaditya

há 6 mesesdiscuss

What struck me is that Vision Pro’s problem isn’t price or hardware, but mental models. Apple keeps framing spatial computing through TV/movie conventions, when the real power is presence with minimal mediation. At least for me, long takes, fixed viewpoints, fewer edits feel “boring” on TV but transformative here.

codebyaditya

há 6 mesesdiscuss

Cowork feels like a real step toward usable agent AI — letting Claude actually interact with your files rather than just answer questions. But that also means we’ll really learn how robust (and safe) this stuff is once people start trying it on messy, real workflows instead of toy tasks.

codebyaditya

há 6 mesesdiscuss

yeah, the job title data is pretty wild. 62.8% of these apps are just 'Analyst' or 'Developer.' From a data perspective, using those generic SOC codes lets high-volume firms standardize everything. It meets the 'specialty' degree requirement on paper, but avoids the higher wage floors that a more specific title like 'Machine Learning Engineer' would trigger. Essentially, the system is being used for scale rather than niche talent scarcity, which shows up clearly in that $50k wage gap.

codebyaditya

há 6 mesesdiscuss

The data shows it’s a structural feedback loop. Even comparing identical job titles in the same city, the $50k gap persists. The gap widens because citizen workers can leave staffing firms for raises at any time, while H-1B holders face 'mobility friction' (60-day rule/backlogs) that keeps them locked into lower-paying tiers longer.

codebyaditya

há 6 mesesdiscuss

Exactly. We call this the 'Innovation Tax.' By legally tying workers to a single employer, the system prevents them from becoming employers themselves.

Our data showed 62% of H-1B filings use generic job titles like 'Analyst'—suggesting that even highly specialized founders are forced into 'cogs in the machine' roles. This isn't just a wage loss for the worker; it's a job-creation loss for the U.S. economy. No taking any side, but based on real data analysis quoting this.

codebyaditya

há 6 mesesdiscuss

Spot on. That 15+ year backlog turns a 'temporary' visa into a long-term economic trap. Our data actually showed the wage gap widens the longer a worker stays on H-1B (rather than converging with citizen wages), precisely because they are locked out of the free market (EAD) for the prime earning years of their career.

codebyaditya

há 6 mesesdiscuss

This is excellent feedback. You are absolutely right that a multivariate regression controlling for location, experience, and job family is the rigorous way to isolate the staffing firm coefficient from the raw data. We stuck to descriptive statistics (medians/distributions) for this initial post to keep it accessible to a general audience, but the 'Cap Exempt' comparison you suggested is a brilliant idea for a validity test. I’ll definitely look into and will try to add a 'Cap Exempt' binary variable to our roadmap for Part 2. Thanks for the 2 cents, it’s worth a lot more than that!

codebyaditya

Submissions

USCIS will weight H-1B lottery by salary starting Feb 2026

New H-1B Rules Hurt Tech Companies 3x More Than the Staffing Firms [OC]

Jobs Paying $250K Get 10x Fewer H-1B Workers Than $200K Jobs [OC]

Where Doctors Are Needed Most, H-1B Petitions Are Lowest [OC]

2.4M H-1B records show $50k gap: staffing vs product companies

comments