I'm still on 1P7 (personal, not team) and still using Dropbox for sync. Hope it never breaks. Never wanted to research the new 1P8 server sync model. Of course I saw that nasty Dropbox unverified IdP account takeover disclosure, so maybe it's no better than whatever 1P is doing on the server.
In our case we use both Dropbox and also have a few teams using the 1P.com sync in addition. So we value both, plus just plain still don't like the 1P8 software itself.
Of course with the original Dropbox based sync, a concern is that Dropbox itself isn't set in amber, at some point it might get rug pulled and sadly 1P never introduced WebDAV or some other self-hosted option before subscription/VC fever set in. And I suppose that macOS continues to move along too. 1P7 on the latest and final version is Apple Silicon native and through a lot of major transitions doing nothing exciting, so it might well keep running for a long time. But the browser extensions will probably stop working at some point which is going to be another big line for a lot of people, and legitimately because it's a not insignificant bit of security.
Anyway, I mean, not like one couldn't keep running it a long time via various layers, but I figure probably time to at least survey the landscape on moving on :(
https://infiniteslop.ai
Generates infinite stream of 10-second vignettes based on user prompts. very curious what the prompt-upsampling prompt is on this one.
Apple is rumored to have released updated models of the Mac Mini and Mac Studio "off-schedule" in part to respond to unexpected local-model driven demand. Hopefully they will invest more in the software side.
"At some point, it raises the question of what actually is the market. If everyone is buying stocks based solely on their market value, how is that value actually being set?"
…
"According to research by Hendrik Bessembinder, professor of business and finance at Arizona State University’s W.P. Carey School of Business, over the 55-year period from 1971 through 2025, an equal-weight strategy beat the traditional value-based index approach by an average of 1.3 percentage points a year, or 12.6% to 11.3%. As the proponents of index funds repeatedly argue, such differences really make a big difference over time."
Right! Maybe the longer into the future the older the kid is. It's hard to predict past 1s sometimes what my toddler is going to do. The older one, I'd bet on the outcome within a couple seconds.
They just haven't re-rendered the map tiles yet. See the hurricane banner here [1]. "Per Secretarial Order 3453, Lake Ontario has been renamed to Lake America. The basemaps for GNIS are in the process of being updated to reflect this name change."
I think the "speed" is because this change probably just shows up in Google by virtue of changes propagating in the GNIS feed [1]. It showed up in the feed Aug 27th with a note on the executive order [2].
[2] 'Secretary's Order 3453 directs the Board on Geographic Names ("BGN") to immediately rename Lake Ontario to Lake America and to update the Geographic Names Information System to appropriately reflect the change. This Order implements President Trump's direction in Executive Order 14420 "Honoring the American History of the Great Lakes and Renaming Lake Ontario as Lake America," that the name be changed. The variant name, "Kaniatarí:io Lake", is a Mohawk name which translates to: "nice lake".'
Yes, the "neural accelerators" in the M5 GPU cores really do 4x pre-fill performance over M4. You can find benchmarks online since M5 has been out for months now. I assume the M6 has whatever the next-gen version of those is. These are not the "neural processors" that have been there since M1 (those still exist) but are additionl matrix math accelerators inside the actual GPU cores. Like "tensor cores" on Nvidia GPUs.
Thank you, found some benchmarks and they look really promising. To my understanding those Neural Accelerators are like AMX but for GPUs. With those accelerators the GPU performance on a M5 Max in LLM inferencing would totally be on par with a 5090, that's quite impressive!
Which inference benchmark are you looking at? The prefill speeds should be comparable on some LLMs, but the 5090 has much a higher theoretical max decode speed.
With larger (also really fast) unified memory on the chip, it could easily load larger weights, and the thing is that the quantized results on the M5s are really good, I believe that for most local inferencing, users would be using INT4 (maybe more bits per weight sometimes) quantized models, that might be where the Neural Accelerators kick in. In raw power, a desktop 5090 easily outperforms the M5 Max, but for this specific use case, I believe that the M5 Max is good enough.
reply