Jon Deutsch

@RichardHanania Unvarnished truth

Every generation of leftists seems to just hate whatever new thing most represents capitalism and is making life better. Then they forget and move on to the next one. First they obsessed over Walmart. Its crime was a wider range of goods and lower prices. Then Amazon was the issue, because it also gave low prices but also brought unimagined variety and convenience. One click order to get anything in the world overnight, sane people would've been singing its praises, but they specifically targeted them. Now it's data centers, the main engine of economic growth. Populist rightoids are joining them. Good thing about the American system is that it's too deadlocked and too subject to special interest influence to wipe out whatever is new. So data centers will make life better, and then they'll hate the next thing.

@mattyglesias Yet are you not a dogmatic ideological moderate? "Moderate" needs to be retired as the frame. Alternatives include: Relevant Pragmatic Innovative Flexible Reasonable

Absolutely! So I would love to ditch “moderate” and work with the people who say that what really matters isn’t ideology but elevating “outsiders” who “fight” and care about “affordability” — the problem is that 90% of them are in practice dogmatic ideologues.

@mattyglesias The trick here is that "relevant" or "novel" or "innovative" gets you out of the Left/Right axis of political thought and allows for truly heterodox ideas. "Moderation" locks you into "less" Left or Right. I think the "less than" orientation doesn't do moderation justice.

@mattyglesias No, it's not moderate - it's relevant! We need to use better words for this movement.

Okay but this is exactly the point — what do you call a Democrat who rather than being a down-the-line progressive appeals to voters who have a less cohesive ideology? That’s a moderate, right? Someone who breaks with the left on some salient issues.

@JonahDispatch Actually all times of the day 😎

At the end of the day, isn’t everything powered by physics?

“Both the impulse to improve and the impulse to conserve are necessary to the healthy functioning of any society. Whether we join our energies to the party of progress or to the party of permanence must depend upon the circumstances of the time.” ~Russell Kirk ᴛʜᴇ ᴄᴏɴsᴇʀᴠᴀᴛɪᴠᴇ ᴍɪɴᴅ

@mattshumer_ @iamjohnoliver This exemplifies @iamjohnoliver's entire shtick on his show - every single bloody topic is handled this way. It's progressive propaganda on par with @FoxNews for conservatives. It's the baseline definition of "truthiness." Entertaining? Sure! Good for society? Nope.

People keep sending me this clip of @iamjohnoliver using my tweet as evidence that AI models don’t work well. Just to clear up any confusion, with respect, the tweet was a) taken way out of context and b) extremely outdated. The model in question (4o) is multiple generations old, and was shut down for being too sycophantic. Current models would not have behaved this way. It’s sort of like looking at a Nokia flip phone and saying “this isn’t useful”, when an iPhone exists. John, I’m a fan, and welcome any discussion here. Just want things to be accurate and not misleading!

@mattyglesias Do you not read @NoahCRothman and @JonahDispatch ?

I find it fascinating that the conservative writers I follow on here and on substack have basically nothing to say — and no interest in — about the subject of why the Trump administration is flailing and unpopular.

@mattyglesias If "taking seriously" = "regulate" then, sure. But regulation is a 20th century tool being misapplied to 21st century challenges.

@MKBHD Nature's Autocorrect

Why does the the brain ignore the second the?

@mattyglesias Bad? Nah. Suboptimal, perhaps!

@mattyglesias The entire conceit and brilliance of democracy is that it's *responsive* -- not proactive. We really, truly don't want charasmaitc ideologues convincing the masses that the sky is falling. On balance, given human abilities, it's better to respond than to predict.

@andrewsharp Excellent article, as always. Hot take: I think you're an even better blogger than podcast host.

@RichardHanania It's Trump vs. Trump

Looks like someone finally found a practical use for crypto.

@underverse @zeynep Cost structures will converge to nominal over time, and offering a "lying is bad" alternative at the high end will find its market.

@jdeutsch @zeynep They can, but they will radically change the experience. Adding confidence checks make queries more time-consuming, more expensive, and less productive because you get more hedging. This makes the tool less desirable commercially. And even then, hallucinations don't go to zero.

It’s incredible how many people — even journalists — who obviously don’t understand much about generative AI will claim the “hallucination” problem have been solved. Soon! Just around the corner. You’re not using the best model! But reality eventually catches up.

@underverse @zeynep They can be added to the architecture.

@jdeutsch @zeynep Guilt and pain are meaningless concepts in the context of LLMs, which have no experience or subjectivity. An LLM can only probabilistically manipulate tokens in response to a query. It cannot self validate.

@underverse @zeynep We need to introduce a new dimension of RLHF (or RLF) that incorporates the concept of guilt and pain for making things up. It's the missing piece in the LLM architecture.

@jdeutsch @zeynep And RLHF can't do anything to help an LLM know when it's making something up, per your original comment. LLMs don't and can't make assessments of their own veracity.

@underverse @zeynep RLHF

@jdeutsch @zeynep This is completely false. There’s no way for an LLM to “know” it doesn’t know an answer, because it is always seeking the probabilistically most desirable answer. It can’t assess its own success internally, by design.

If Pakistan wasn't on @Kalshi yesterday as a reason for DJT backing off his civilization-ending threats, then is it really a prediction market?