Hold2

Build. Share. We invest intelligently in ourselves so that we may share with the world. | Owner - Clayton P. Cobb (@warrtalon) |

[Hold2 @Hold2LLC]

Thread: Masks and NPIs.

This thread reviews information published from Oct 2019 - Dec 2020.

The point is to see what was known/believed before & after discovery of COV2.

We'll start with research on epidemic/pandemic influenza (flu) from Oct of 2019.

Aerosols vs. droplets.
/1

[Hold2 @Hold2LLC]

@artvand7799 @SuperGrok That sums it up.

[Hold2 @Hold2LLC]

@SuperGrok I mean... really? This is as far as AI has evolved? Incessantly gives the wrong answers on basic calculations. Even when it shows itself its own mistake, it doesn't believe it and sticks to the original wrong "verified result." It then takes me 5 more messages to break it loose, but then on the very next request it does it again and acts like it has no idea we ever had the discussion.

[Hold2 @Hold2LLC]

Is this true for all LLMs? Is there a version of any LLM where this type of idiocy doesn't continually happen?

I get this all the time with SuperGrok, ChatGPT (paid), and GitHub CoPilot (all agents)

TLDR of the screenshot is:
- I ask a question related to the immediately-preceding exchange (most recent 4-5 messages)
- @SuperGrok gives the wrong answer by 1 message
- I tell it that's the wrong answer
- It bakes the wrong answer into every reply no matter how much I scream until I tell it to literally print the conversation, after which it goes, "Oh, oops, you're right."
- Its excuse is "persisted reasoning"

[Hold2 @Hold2LLC]

@AlexZwitter2 I concur with and have learned that the hard way.

What I am frustrated with here is that it can't recall facts from the recent conversation history...or rather, it doesn't bother to actually check the history before answering and then bakes in that wrong answer permanently.

[Hold2 @Hold2LLC]

Is this true for all LLMs? Is there a version of any LLM where this type of idiocy doesn't continually happen?

I get this all the time with SuperGrok, ChatGPT (paid), and GitHub CoPilot (all agents)

TLDR of the screenshot is:
- I ask a question related to the immediately-preceding exchange (most recent 4-5 messages)
- @SuperGrok gives the wrong answer by 1 message
- I tell it that's the wrong answer
- It bakes the wrong answer into every reply no matter how much I scream until I tell it to literally print the conversation, after which it goes, "Oh, oops, you're right."
- Its excuse is "persisted reasoning"

[Hold2 @Hold2LLC]

@el_smurf0 Yeesh. That sounds like exactly the same root issue.

[Hold2 @Hold2LLC]

@VotesForLizards Yes, when asking @SuperGrok to explain why, it said,

"I did not 'review the history to be sure' in the way a human would; I generated claims based on a corrupted internal summary that persisted across turns."

It also claimed this is a well-known limitation of LLMs.

[Hold2 @Hold2LLC]

@ChauFITOfficial @wrb7271 @DrNeilStone Yes, basic logic, but you said “deaths went down.” As with Bill, I’m just correcting the specifics.

[Hold2 @Hold2LLC]

@ChauFITOfficial @wrb7271 @DrNeilStone C19-associated deaths increased at a similar pace after mass vaccination.

IFR and CFR likely decreased, per capita deaths increased dramatically in countries that had fewer cases before vaccine availability.

Here are some grouped examples of both types of countries.

[Hold2 @Hold2LLC]

@ChauFITOfficial @wrb7271 @DrNeilStone I have all those charted, too. Bill’s claim was wrong, and many people still hold his misconception as true.

One major issue with comparing Vax vs. Unvax rates is the Unvax denominator is never known. The discrepancy between UKHSA and ONS was ~6 million.

[Hold2 @Hold2LLC]

@ChauFITOfficial @wrb7271 @DrNeilStone The reason I put all the red boxes with qualifiers and bold text was so people would not respond like you did, but I can never seem to put enough qualifiers.

Please look at the tweet I was responding to and how I responded.

[Hold2 @Hold2LLC]

@wrb7271 @DrNeilStone This is a common misconception. The vast majority of COVID deaths were and are vaccinated. That’s because most deaths are elderly, and the vast majority of elderly are vaccinated.

A small percentage of COVID deaths are unvaccinated, and that % decreased over time.

[Hold2 @Hold2LLC]

@ZacBissonnette That linked piece you mention is incredible.

I wonder if the person who wrote that really believes any of it.

One egregious example here.

[Hold2 @Hold2LLC]

@akm515 We’re double that in NY in addition to the 2nd highest income tax.

[Hold2 @Hold2LLC]

@sgillern @ClownBasket @FrankLuntz Correct. Per @nycHealthy

[Alessandro Palombo @thealepalombo]

8. Why Trump’s Pick for NIH Matters

Dr. Jay Bhattacharya was one of the few voices to challenge the government’s Covid policies.

Benz explained why Trump naming him as Director of NIH is so important.

[Hold2 @Hold2LLC]

@othingstodo_com Yes. I wore a key around my neck to/from school at age 8 and had one rule to live by: be home before dark or be grounded.

[Hold2 @Hold2LLC]

@odinikaeze @JoshKraushaar

[Kelley K @KelleyKga]

Okay, I've got to admit... This meme made me smile this morning. 😁 Honestly, I'd love the opportunity for a role at the CDC to help public health rebuild some trust, and I'm already in Atlanta...🤷‍♀️

[Hold2 @Hold2LLC]

@DrClaytonForre1 @kalezelden Roll Tide! I’m from Alabama but was shocked to see how many kids here on Long Island (NY) are going to UA. It’s awesome to see.

My son also was recruited actively and offered scholarship money without even applying for it. He’s deciding which southern school to attend.