Your question makes me think of the last scene in the movie Night of the Comet.

I think the 1.76 includes that. Plain ternary packed into bytes would be around 1.58.

When it can imagine real risks and not hallucinations, I am sure it will be better than having a human lawyer write the agreement. I am not convinced that we are there yet.

>"Information wants to be free".

I agree wholeheartedly and in keeping with that, I call upon frontier AI labs to release both their weights and training sets.

True, eradicated is a strong word and I do think it should be. And as horrible as a cataclysm is, I think it's still going to be less pain for humanity than going forward with this tech.

Are you seriously putting Balzac and Spiderman on the same level to try to ridicule thoughts?

Ad hominem.

JEV is a new type of AI (https://typesafe.ai/blog/introducing-system-one-models-and-j...)

I thought the Kiwi API was invite only for a while?

Anyway, looks cool - and seems very fast!

That makes it even worse -- spending that much time with AI should make it glaringly obvious how valuable it is.

It’s just another reason for companies to say they can’t find skilled labour so they can bring immigrants and underpay them.

Berkshire returns cash to shareholders through buybacks now. It is more tax efficient. And no they do not speculate - they invest.

using them with 1password has been pretty effortless. 'want to add the passkey' sure why not. 'want to use the passkey?' sure why not. for me it works across devices/os/platform so not sure what the big gripe is tbh.

It might be conceptually similar to a single-output-token LLM (sort of). LLMs output next-token probabilities. You can ask LLMs to output yes/no, or to output only a color, or only a digit or something like that.

In this case I would imagine that they probably embed your input data into a vector space, and they embed your questions/outputs into another space, and manage to predict probabilities/classes/scores for your outputs very quickly. Embedding the output classes/questions into a vector spaces gives you something you can reuse across runs cheaply, as opposed to an LLM where you can prefill the KV cache but this is an expensive operation in terms of memory.

Google has been scraping everything from us since day one. Meta, Microsoft, Github, Slack, Reddit, StackOverflow, big and small, every single app that interacts with people uses our own data to make money and create walled gardens. I haven't seen a single one opening their silos to the world. That's our data, we produced it, you captured it and now you think it's yours

So no, your cries for regulating others because you are losing the race won't work this time.

The unslopify toggle is pretty funny though lol

Thank you! Perhaps I should add the supported functions in README.md. There is a list in programme's help (F1). I copy from there:

ABS Absolute value AND TRUE if all arguments are true AVERAGE Arithmetic mean of values AVERAGEIF Mean of cells matching a criteria COLUMN Column number of a reference COUNT Count of numeric values COUNTA Count of non-empty values COUNTIF Count of cells matching a criteria COUNTIFS Count matching multiple criteria pairs DATE Date serial from year, month, day DAY Day of month of a date serial DAYS Days between two dates EXACT Case-sensitive text comparison EXP e raised to a power FILTER Filter array rows/cols by include booleans with optional if_empty FIND Position of text within text HLOOKUP Horizontal lookup in first table row IF Value depending on a condition IFERROR Fallback when an expression errors IFNA Fallback on #N/A like IFERROR IFS First matching condition's value INDEX Value at row/column inside a range INDIRECT Reference from text (A1 or R1C1, with sheet) INT Round down to integer IPMT Interest portion of a loan payment ISBLANK TRUE if the cell is empty ISERROR TRUE if the expression fails ISNUMBER TRUE if the value is a number ISTEXT TRUE if the value is text LEFT First characters of text LEN Length of text in characters LET Bind names to values for a calculation LN Natural logarithm LOWER Text converted to lowercase MATCH Position of a value in a range MAX Largest numeric value MID Substring by start and length MOD Remainder with sign of divisor MONTH Month of a date serial NA The #N/A error value NOT Logical negation NOW Current date and time serial OFFSET Range offset by rows/cols with optional height/width OR TRUE if any argument is true PI The number pi PMT Loan payment for fixed rate/terms POWER Number raised to a power PPMT Principal portion of a loan payment PRODUCT Product of values RIGHT Last characters of text ROUND Round to given decimals ROUNDUP Round away from zero ROW Row number of a reference ROWS Number of rows in a reference SQRT Square root SUBSTITUTE Replace occurrences of text SUBTOTAL Aggregate ignoring nested SUBTOTALs SUM Sum of values SUMIF Sum of cells matching a criteria SUMIFS Sum matching multiple criteria pairs SUMPRODUCT Sum of products of corresponding array elements TEXT Format a value as text using a number format TEXTJOIN Join text with delimiter, optionally ignoring empty TODAY Current date serial TRIM Strip and collapse spaces TRUE/FALSE Boolean literals UPPER Text converted to uppercase VALUE Convert text that looks like a number to a number VLOOKUP Vertical lookup in first table column WEEKDAY Day of week of a date serial XLOOKUP Modern lookup with match modes YEAR Year of a date serial

Regarding the 2nd part of your question, I quote from the LLM: Complex scalar formulas with deep, cross-sheet, and indirect dependencies are fully evaluated with cycle safety; range/array inputs compose inside aggregations, but dynamic-array spilling is not implemented — multi-cell results surface as #VALUE! instead of spilling.

It's nice to see some variation in websites. Makes each one feel so special.

Thanks. I'm upvoting that instead

I am conflicted with passkeys.

I actually prefer non-resident U2F in some ways. You don't have to store anything on your key, you are just signing requests. This is relevant where U2F/FIDO keys have limited slots for 'resident' keys.

In principle, it's great. You have one good password to remember for the average user, and that's enforced by their device's probably good enough security posture.

They are resistant to being phished and they won't reuse the same one everywhere. They then don't end up going from hunter2 to hunter2! everywhere.

But my experience for users is that they worry they are giving their biometrics to Amazon or whoever and so the UX just confuses them.

The certification aspect was new to me too last time passkeys came up. Sites can require that a given passkey has been certified.

The patchy support for them is also frustrating. MacOS does not support NFC FIDO/U2F. iOS does.

I have this weird vision of an alternate reality where governments (say, National Archives) are the ones creating the models as a public service and then the rest of the industry is just commoditized pricing of hosting them, competing with value add bits. And we’re on here reading articles about how the latest release of the EU model does a better job generating maps now and the new Canadian model seems to apologize less.

I used Authelia in the past, I'm curious: why not using Cloudflare Access?

Knowledge is neither inherently good or bad.

I've resisted passkeys for some of the reasons listed. I use third-party password managers, they're device/system independent, and I can export the data when I want to. I don't see passkeys as a big advantage, or perhaps I should say the bigger advantage isn't really for the _user_.

Of course, at the rate we see security failures everywhere, I'm not entirely convinced writing your passwords on post-it notes wasn't such a bad idea after all.

Not necessarily as there are methods to redistribute wealth without the consumers having any choice.

Recently, there was the example of the SpaceX IPO listed on Nasdaq (after they changed the rules to allow it). Lots of people have pensions/investments in tracker funds and those funds are essentially forced to buy SpaceX shares.

Simple things like "quantitative easing" can result in higher inflation which essentially devalues people's money. The ultra wealthy will typically not have any meaningful percentage of their money in currency, but instead will be in various assets around the world which means that their wealth is not affected by the inflation.

There's plenty of other schemes such as the "too big to fail" method of securing handouts from the government.

Not always. And which vault? There can be multiple on a given device. This isn't some hypothetical 'mollify the user's worries' question, this is an important practical question of what do they need to worry about losing access to. Trust me when I say that most users I have talked to have absolutely no idea about this, and usually only find out when they've already lost them.

chinnese doing chinnese thing

Won't missing bands affect the reliability as well, not just the speed?

Like…?

It depends on what you called solved.

If the goal is to only extract the unstructured text from the document, it is definitely solved. Extracting a more natural structure like paragraph separation, tables, header, footers (what is referred as document intelligence) is much more complicated and not fully solved, but I would say almost.

> Polynesian languages would distinguish between fresh water bodies and salt water ones

Māori distinguished water _quality_ without reference to whether or not it was a lake or sea or river.

Wai tai: Salt water

Wai māori: Fresh water, water for regular daily usage

Waiora: Pure water that restored/healed a person

Waikino: Dangerous / polluted water

Waitapu: Sacred water

Which makes sense right? You can have a river that is significantly salty and hence unpotable some distance from the sea, depending on the strength of the tides - likewise you can have a lake near the sea that intermingles with seawater to the extent that it's brackish.

So Māori naming was more based on utility than geographical distinctions developed in Europe.

Have you considered that maybe, just maybe, you can have both an easy path to your paycheck and care about software engineering? You also might have an incorrect definition of "software engineering" in mind