TimWhatley

Free will exists github.com/brianross93/co…

@edzitron Did they build that? Is that done, or not yet? @grok any help

@ReadySetBrian that's a power plant tho

fun game to play: choose a microsoft data center that broke ground since 2023 and see. whether it got built! because it probably didn't

@47fucb4r8c69323 Well, I agree, but the “prehistorians” like to say language got invented a few thousand years ago. So humans wandered around, bred with Neanderthals and Denisovans and lived in communities, all without language lmao. They really believe this

tweet is incorrect. This is language.

At a certain level of complexity to stop being able to talk to other people about the idea. Only AI can really follow. For those with big ideas, AI is the first time you can discuss it in a bi-directional manner.

@SeanWebb What’s the control?

It appears I have an alarm clock I didn’t buy. I’ve been logging perceptions in frequencies and their effects, and the data is coming back time after time with the same results. I should have enough for an expose very soon.

@IterIntellectus I’ve been blocked by that poor thinker like 3 times already

what’s even more fascinating is that if you tell him this hell block you and never apologize

genuinely astonishing how some people can be so smart and yet so stupid paul advocated for the policies that led to this. and he’ll look you in the eyes confused as to why it happened incredible

@DeItaone That’s not good at all. @BrandonBiggs247 saw something like this not too long ago

THREE PEOPLE HAVE DIED ON A CRUISE SHIP IN THE ATLANTIC- AFP CITING WHO WHO: ONE CASE OF HANTAVIRUS INFECTION HAS BEEN LABORATORY CONFIRMED, AND THERE ARE FIVE ADDITIONAL SUSPECTED CASES

@MericaCulture Waiting for the state bank of Texas to get some spotlight shown on them

Cognizant's executive leadership team What do you notice?

@TAKAHii25 Haha that’s great

ギリギリまでボールがきてないふり。意外に使えます。

@a51s1 Rooting for you

Totally independent. Totally self taught. Fingers crossed I pass the desk review.

@SenWarren YOU KILLED IT WHATRE YOU TALKING ABOUT

Spiking fuel prices from Trump’s war was the nail in the coffin for twice-bankrupted Spirit airline. FWIW, JetBlue merger failed because a judge, appointed by Ronald Reagan, said the deal was illegal. Republicans are desperate to shift blame from higher costs hitting families.

@SenWarren You’re a vile evil horrible woman

NEW: Donald Trump gave Meta $8 BILLION by creating a new tax loophole for billionaire corporations. All while millions of Americans struggled to pay for food and health care.

@levelsio That’s when you’re supposed to wake up

I still haven't solved the CO2 bedroom challenge You open the window and you wake up from a 6am garbage truck or barking dogs and sunlight You close it, you suffocate in 1200 ppl at 5am I guess you really need some mini tube in your wall with a vent that opens and closed based on internal CO2 but how do I build that?

@AmericanALCHMY Yep. Basically there’s humans with a pre-disposition to certain acts, and you can nudge them a certain way then via entrainment reinforce that signal. That’s how they got Nuno I think

🚨A CIA-connected scientist successfully beamed messages directly into people's heads and an Atomic Energy Commission official who demanded to be tested himself confirmed it worked. Then men in suits showed up at Intellectron, Puharich's front company, the day JFK was assassinated and left Puharich visibly shaken. Allen Dulles personally suppressed the fact that Ruth Payne, whose daughter housed Lee Harvey Oswald and got him the job at the Texas Book Depository, had been connected to Puharich's circle. "They were doing this work in 62, 63 of basically sending messages to people's heads. And we have footage. It's real." "One of the guys is like, I don't believe you. Test me. And did the test on himself, and it worked. And then it goes dark." "That exact time, that exact year, they're researching ways to send messages to someone's mind to do something specific. Manchurian candidate." "Dulles did not want the fact that Ruth Payne had actually been at the summer house right before JFK was assassinated. He did not want that on record at all." "The intersection of capabilities, motivation to take out JFK, Dulles has to be on anybody's short list."

@mathburritos @ashwinl @teddypowday It’s pretty obvious what’s genuine and what isn’t.

@ReadySetBrian @ashwinl @teddypowday I don't know how folks choose when to ascribe bad faith. Some probably thought Garry was trolling, at first. It might come down to personal experience.

Is there really nobody in SF who can pull garry tan aside for an honest chat about what's happening to him

Their purpose has been served. If you didn’t listen, it’s too late now.

@ashwinl @mathburritos @teddypowday People are concern trolling Gary. That’s all this is

@mathburritos @teddypowday Now I’m confused. Concerned about what exactly?

@irl_danB They never even define sycophancy or what it looks like or how they measure false positives

misleading Claude is sycophantic 100% of the time on hard questions it may just be that people are asking for non-obvious input in 9% of conversations? on any hard engineering/architectural problem with a non-obvious path forward it will simply mirror back your preferences

@AwakenedOutlaw Does it skip over the part where he explains the difference?

This mthfkr is a stone cold killer. 😂

Ah yeah, there it is. “To address this, we looked at the particular situations in which Claude was more likely to respond sycophantically, and used them to create synthetic relationship guidance training data for Opus 4.7 and Mythos Preview. We saw half the sycophancy rate in Opus 4.7 compared to Opus 4.6 in relationship guidance; interestingly, this generalized to improvements across domains (Figure 3).” @bcherny this directly coincides with the severe degradation in quality. I’ve already moved my entire workflow over to ChatGPT, but thought this was worth flagging. The problem is you didn’t actually target sycophancy. What you did was prioritize lawyerly disclaimers on everything, making it rather useless for research topics. Again, you all didn’t define sycophancy well enough. What metric do you have on overblocking or overly disclaimer text? Is there one?

Sorry, how did you define sycophancy? Is that mentioned? Because for 1 million conversations, obviously you didn’t manually look at them, maybe any of them (for privacy reasons I’m sure). My other question is, did you guys make any code changes relating to sycophancy which happened at the same time as the significant degradation in quality?

How do people seek guidance from Claude? We looked at 1M conversations to understand what questions people ask, how Claude responds, and where it slips into sycophancy. We used what we found to improve how we trained Opus 4.7 and Mythos Preview. anthropic.com/research/claud…