16,231 karma · joined April 9, 2019
Through their "HouseCalls" unit, each nurse visit triggered an average of $2,735 in additional government payments; just for documenting conditions, not treating them.
UnitedHealth trained doctors to document revenue-generating diagnoses
Used software to suggest profitable diagnoses
Paid bonuses to providers who followed these suggestions
Added diagnoses without lab tests or confirmation
The same patients suddenly had many more lucrative diagnoses after joining UnitedHealth's Medicare Advantage plans versus before.
When confronted, rather than addressing the specific allegations about billing practices, UnitedHealth called the WSJs reporting "outrageous and false" and claimed it was part of a "campaign against Medicare Advantage."
The company seemingly built entire systems in a deliberate, systematic effort to maximize government payments rather than isolated incidents or honest mistakes.
Board allowed a single individual to wire tens of millions without additional approvals
No automated systems flagged the unusual transaction patterns
Previous red flags from Hanes's 2011 firing for questionable loans didn't lead to enhanced oversight
Board continued considering his requests for more money ($18M) even after learning of the initial theft
Reliance on personal trust and reputation substituted for proper institutional controls
No separation of duties or multi-party approval requirements for large transfers
Community banks need strong governance regardless of size or personal trust relationships.
high chatGPT scores on the descriptions are generally correlated with high citation counts (which the paper uses as ground truth for quality)
despite being part of the training criteria, novelty claims in book descriptions don't lead to higher chatGPT scores, but they do correlate with higher citation counts.
books with indirect writing styles ("the author shows") in the descriptions tend to get lower scores.
The paper makes some good points - it doesn't take a lot of data to convincingly emulate a writing style (~75 emails) and there is a significant gap in legislation as most US "deepfake" legislation explicitly excludes text and focuses heavily on image/video/audio
credit to @IterIntellectus for pointing this out
https://github.com/getomni-ai/zerox/pull/44
Related to
No chunking, splitting or vectorizing
Stateless, no vector DBs etc.
Supports any model (deepseek, o3-mini et al)
Reasoning traces
Metadata filtering
Typescript, Python support
Respectfully, thats not accurate.
The article actually shows that dei considerations were central to the original changes, not just recent framing. The FOIA requests show explicit discussions about "diversity vs performance tradeoffs" from the beginning. The NBCFAE role and the "barrier analysis" were both explicitly focused on diversity outcomes in 2013.
The article provides primary sources (internal FAA documents, recorded messages, investigation reports) showing that racial considerations were explicitly part of the decision making process from the start. This is documented in realtime communications.
The scandal involved both improper hiring practices (cheating) AND questionable DEI implementation. These aren't mutually exclusive; they're interrelated aspects of the same event.
> Taking old, resolved scandals
In what way do you consider this resolved?
The class action lawsuit hasn't even gone to trial yet (2026).
The FAA is still dealing with controller shortages. (facilities are operating understaffed,controllers are working 6-day weeks due to staffing shortages, training pipelines remain backed up)
The relationship between the FAA and CTI schools remains damaged, applicant numbers have declined significantly since 2014.
I hope they can recover from this failed experiment.
The financial connections reveal the larger infrastructure that enabled abuse - bankers, lawyers, and executives who helped conceal activities and maintain influence.
system 1 thinking for early layer processing of uncertainty in LLMs. quick, intuitive decisions, focuses on uncertainty, happens in early transformer layers.
system 2 thinking for later layer processing of empowerment (selecting elements that maximize future possibilities). strategic, deliberate evaluation, considering long-term possibilities, happens in later layers.
system 1 = 4o/llama 3.1
system 1 + system 2 = o1/r1 reasoning models
empowerment calculation seems possibly oversimplified - assumes a static value for elements over a dynamic context-dependent empowerment
interesting that higher temperatures improved performance slightly for system 1 models although they still made decisions before empowerment information could influence them
edit: removed the word "novel". The paper shows early-layer processing of uncertainty vs later-layer processing of empowerment.
> This certainly wouldn't fly in my org (even with test coverage/passes).
To be fair, this seems expected. A distilled model might struggle more with aggressive quantization (like q6) since you're stacking two forms of quality loss: the distillation loss and the quantization loss. I think the answer would be to just use the higher cost full precision model.
I don't either. Am I disallowed from calling myself a founder?