celestialcheese

Does Signature failing mean crypto in the US is finally dead? Silvergate and Signature both going to $0 is a strong signal.

Why would any bank want to work with crypto after this?

For sure. If perfect accuracy is important, it's still good to do sampling and human review to figure out accuracy rate and decide if further checking is required. But it still beats the pants off accuracy of other methods for the amount of work required.

That said, with fine-tuning, `davinci-003` is _excellent_ at the types of entity extraction you're describing.

1000x this. Entity extraction from unstructured text with zero/few-shot is fantastic.

I've got a use case where I need to extract model numbers from text - these LLMs are so good at it with very little work.

Asking honestly - if you just got $100m wired to your account from a Series C, what's the right way to protect your cash?

Where are you getting 170b in shortfall?

SVB has ~$200b before today. 30% of Lehman, but not nothing.

Was this made for the shelved HBO series?

Most of the critique I hear Stephenson get is that he writes an incredible 400 page book in 800 pages.

I agree with you, though. His world building and stories pull me in unlike almost any other author - the more the better IMO.

How does it compare to DeepL? And per-word human translation services like you'd source through Crowdin?

Here's one anecdotal source confirming this - specifically some copywriting contractors. Significantly less hours needed per month writing. A lot more hours for our editors but it's still significant cost savings when netted out.

I've found that context is everything to getting consistently good output, and augmenting your prompts with known truths with SERPAPi, embeddings and a vector db, brought really flakey results into the >90% accuracy threshold.

As an aside - does anyone have good tools or methods for testing and evaluating prompt quality over time? Like performance monitoring in the web space, but for prompt quality. The techniques to use LLMs as evaluation tools of themselves always has seemed flakey when I've tried it, I'd like to use a more grounded baseline.

For example, if you have a prompt that says "What is the weather today in {city}?", you can run it against a list of cities and expected outputs (using a lookup to some known truthful API). That way, when you make changes to the prompt, you can compare performance to a baseline.

The next sentence of that quote is pretty illustrative.

> Pathmatics said the previously unreported figures on Twitter advertising are estimates. The firm bases its estimates on technologies that track ads on desktop browsers and the Twitter app as well as those that mimic user experience.

> But the company said those estimates do not account for deals advertisers may receive from Twitter, or promoted trends and accounts. “It is possible the spending data could be higher for some brands” if Twitter is offering incentives, Pathmatics said in an email.

It's all speculation across the board - people want musk to finally fail, and that possibility produces some delicious schadenfreude. I am in this industry, and don't trust for a second the estimates "Pathmatics" cites. Everytime I've seen these types estimates on properties where I know the real numbers, they are off significantly, in both directions.

The numbers come from comps, and some sampling of Desktop ad impressions and a twitter client. That's so far away from what the real numbers are that it's just a fuzzy guess.

1 - https://www.euronews.com/next/2023/01/19/twitter-musk-consum....

It's almost as if the media narrative around Twitter might not match reality.

This is false.

https://developers.google.com/search/blog/2023/02/google-sea...

Unless "Steadipulse.com" is now the new canonical source for Google's webmaster guidelines, AI content is fine as long as it's human reviewed. Spam is still spam, and they consider unreviewed/unedited AI content to be untrustworthy.

This is a new one. What do you think pensions are?

So your hot take is that retired government employees, union laborers and other beneficiaries of pensions deserve to have their contribution to the pension and their retirement income zero'd out just because you don't get some of the money?

Who's going to pay for the repurposing? Converting an office to residential or lab space can cost between $200-$350/sqft in west coast downtown locations.

Not necessarily - I don't know the details of these security providers, but they could use a proxy scraping service like BrightData, which proxies requests through real users mobile and residential IPs, thus making it look like real users are clicking.

I don't know this exact thing to be happening, but it could, so I don't think using IP location as the signal of "Human vs Not" is accurate.

Using time clustering of clicks vs send is a good one, also I've noticed that these security providers will click _every_ link in the email, all within a few minutes, so filtering out multiple clicks by one IP in a short time frame would work.

20% was a showstopper when cities could drag things out for 10 years with red-tape and consultant hell.

Now it could be doable with a lot of that red tape is gone.

No, but people are mobile, and the houseless are also people who communicate through the internet and share where assistance is available.

If free, safe lodging is available without restrictions, demand will quickly outstrip supply with folks moving from other cities (LA, Seattle, Portland, etc) to fill the available rooms. It's a $50 bus ticket, city officials will likely fund these tickets for them to get to SF to take advantage of this. SF had programs for this exact thing to get homeless out of SF [1]

1 - https://www.vice.com/en/article/bvg7ba/instead-of-helping-ho...

When done right, it works.

It's the hubspot/marketo approach of slamming inboxes with daily/weekly, non-specific garbage that needs to go. And buying lists or using lists from conferences - fuck that, it's just spam.

Following up with someone who explicitly gave you permission to send them emails after signing up for your service is reasonable and expected IMO.

It's the only channel that isn't owned by a walled-garden, so investing in it and treating it with respect is essential since it's the only way to communicate to customers and potential-customers without a middle-man extracting rent.

This.

Additionally, many enterprise email security providers doing automated phishing checks that click every link in the email and screen scrape the landing page to verify it's not an attack vector. If your market is the govt, schools, or large enterprises, this is absolutely juicing your CTRs.

Vanity email metrics are incredibly broken right now - the only numbers that mean anything are frictioned conversions like sign-ups, payments, session duration etc.

I don't know about pandemic checks causing this, but VC's have a huge amount of undeployed capital from the boom of 20/21. https://www.theinformation.com/articles/venture-firms-290-bi...

There's a lot of money specifically allocated to venture funds that needs to be deployed.

> Choosing not to treat someone because "they may contribute to society", even if they are suffering and wish not to suffer from their treatable disorder, is always unethical imho.

I agree - it's absolute misery on the individual level and relieving that pain is a good thing.

The real ethical debate comes when we have the ability to eliminate these conditions through germ-line edits. A rogue doctor in China already did this by making the CCR5 gene non-functional in embryos and the children were born.

The doctor went to prison, and the scientific community roundly condemned his work, but this isn't going away. "Fixing" major genetic diseases in the embryo is possible, and it's a pretty slippery slope once we start down that path.

Preface this by saying I don't know how I feel about this line or reasoning, but this is the ethical arguments I've heard against "curing" and/or preventing psychiatric disorders (ASD, Schizophrenia, Bipolar, ADD/ADHD, etc.)

The argument goes - These disorders are incredibly taxing for both the individuals who suffer these conditions and for their family, friends and colleagues. There is real suffering and pain, and alleviating this pain, and potentially preventing it through germ-line edits could eradicate these conditions for ever (glossing over the incredible ethical concerns there)

At the same time, these conditions _can_ provide incredible gifts to both the individual and society as a whole. History is littered with famous artists, scientists, etc., who have all made important and unique contributions to the human experience. Would they have been able to do these things without their condition? In curing the disease, are we missing these non-linear thought patterns that are common with these disorders? And what cost does that have for society?

Here's the source study since the NYPost is garbage - https://www.nature.com/articles/s41380-023-01959-7?utm_mediu...

Seriously. This is something the NIST is perfectly set up to do.

That's incredible, I love stories like this. Google busses had open wifi?

With enough residential and mobile proxies, you can scrape almost anything.

Recently, the Chinese/Mexican Cartel laundering operation had been operating at 1-2% fees [1]. This was new and historically you're right, but the amount of money wanting to leave China freed up cash to drive down fees.

1 - https://www.propublica.org/article/china-cartels-xizhi-li-mo...

Everytime a story like this pops up, I think of patio11's argument that "the optimal amount of fraud is non-zero". [1]

I work with our local food bank, and got in a discussion with a family member who was shocked at there being no means testing or proof of need to participate. I'm certain there are people we serve who are over whatever threshold would exist, but the cost of excluding someone who's hungry because they couldn't print a paystub or have access to the right documents is _so_ much higher than any spillage from fraud.

With certain aspects of society, such as with social-saftey-net benefits, when the costs of exclusion impact those teetering on the edge, "non-zero" fraud is absolutely the optimal amount.

1 - https://www.bitsaboutmoney.com/archive/optimal-amount-of-fra...