dmitrygr

OMG someone used our data to make an MK model! Just like we did to every author in the world!

Firefox on macOS. site reported M1 and 16GB. Mac is M4 + 24GB

The *architecture* is RV32I_Zmmul which is unambiguously neither new nor custom.

What you describe (the specific implementation details of a core) is *microarchitecture*. In this case clearly a lot of work was done and it is cool, but the *architecture* is indeed RV32I and not custom

I was not undermining anything. I was helping others find the info I sought and took a while to find. This is why i left my comment. Everyone has their own interests. As an example, my thoughts were "whoa... a new architecture... did they write a new compiler or rewrite doom in assembly?" and for that "it is rv32i" would have been a quick answer.

For anyone else curious: custom FPGA-based CPU code but not a custom architecture. This is an RV32I core. The article takes along time to give up this fact.

PC? what pc?

Every single NES game or CGB game would fit on a floppy. More or less all GBC and SNES games could compress down to a floppy

Until relatively recently in computing history, games (and all other programs) were small. Hell, Pokemon Gold fits un a floppy trivially compressed, together with a working GBC emulator. I can trivially fit rePalm unto a floppy and leave enough space for a great PalmOS RTS - "Warfare Inc"

Reductio ad absurdum

Because most of those people are not repeat violent criminals? We only need to incarcerate repeat violent offenders to reduce crime noticeably. This is not controversial. The data is linked-to above.

Wouldn’t you say that the lowered crime rate enjoyed by the innocent as a result is an improvement?

A LOT of crime can be solved. A huge percentage of perps are multi-repeat perps. Putting them away permanently would solve a lot of crime.

"75% to 83% of released prisoners are arrested for a new crime" https://bjs.ojp.gov/library/publications/2018-update-prisone...

You still did not learn the lesson? Once you take possession of a fixed-function appliance, NEVER EVER EVER take any updates, and do not connect it to the internet (CarPlay does not require internet connectivity in the car). Do not buy fixed-function appliances that require internet, that is what computers are for.

So... an overpriced RGB keyboard? That is what we're doing today?

but 512 MB of RAM were too little to run a modern browser, sadly

want to make clock -> need browser...

i do not want to live on this planet anymore

Last few zotero versions clearly have some O(n^3) algorithms in their Word plugin, since any document with > 100 references makes a very fast apple silicon computer slow as hell, "updating references". Killing zotero fixes it instantly (at a cost of un-updated references). I suspect if any testing is done, it is done with 1 or 2 references in a file, and 2^3 still ain't much

Interesting reporting, marred by obvious llm-slop-sounding writing. "You are not building..., you are ..."

this has been fine working since the 80s

  pc1> tar c <SRC_FOLDER_OR_FILES> | nc -l 12345

  pc2> nc <IP_OF_PC_1> 12345 | tar xv -C <DST_LOCATION>

Nope. I don’t want to work on systems that self-handicap in dumb and predictable ways. Mistakes happen. But mistakes that were preventable and knowable at design time … they are inexcusable, and RISCV is chock-full of those.

False dichotomy. They are also quite possibly aqui-hires which then put the good engineers unto actually useful projects.

The author of this is one of the greatest minds of our time. While doing this is cool, he also designed the Hazard3 core in the RP2350 as well as the QSPI unit in it -- the only memory-mapped QSPI unit I've encountered so far that I've not been able to crash or hang.

When the top perf per watt or perf per MHz machine is RISCV, we'll talk. Until then, lol.

The original paper quoth:

our reference-counted translation model, where every variable is pessimistically wrapped inside a Rc<RefCell<T>> type, checks that would usually execute at compile-time are shifted to run time, degrading performance.

The challenge is that

Why is that a challenge? As long as they are open about this, all is OK.

And what happens when your "tests" are also vibecoded. Right now all of these houses of cards reset on human-written tests. What happens without them?

Just wait a year. It will die down just like crypto and NFTs did. Hype goes in cycles.

with knowledge specialization becoming unnecessary,

I will take the other end of this bet

Repeated nuisance alarms have the same effect on all humans, not just on pilots - it trains them to ignore the alarms. Eventually this will lead to non-nuisance alarms being ignored and lives being lost.

This sort of nonsense is well studied in aeronautical world, and will lead to too many alerts, which, in turn, lead to predictable outcomes: https://flightsafety.org/asw-article/normalization-of-devian...

I've had now-ex friends send me unsolicited ai-slop catch-up messages summarizing their life "since we last spoke".

Prediction: absolutely nothing will happen about resolving this. Ever. And eventually they will find a line in the ToS allowing this and point to it — some paragraph that says they may cut any line of service with no notice at all you you still owe them your firstborn and the payment for the rest of your contract, and also you now need to change your name to Verizon.

Meta’s rate limit is ridiculous. [...] feature [...] doesn’t use Meta’s servers. It runs on-device, using the chips inside the glasses [...] I turned off my internet, and it kept working.

BMW heated-seats style move. Classy.