Casteil

>you are getting health insurance, which has a lot of value.

So much 'value' that millions of Americans still end up in medical debt, with many having to resort to medical bankruptcy even as 'insured' individuals.

Too many exclusions, too many ways to deny coverage. The private health care system of the US is broken.

100%

Real question is whether they'll start making their own 'AmazonBasics' branded AI-generated dogshit, just like they do to undercut/displace successful 3rd party products.

Ubiquiti U7 Pro is only around $200 USD.

I just use Ollama[1] - makes it incredibly easy to get going on MacOS, and can be run on linux/WSL also. RAM required will depend, but generally to run Mixtral at reasonable quantization levels (e.g. Q4) you're going to want 36GB or more.

[1] https://github.com/jmorganca/ollama

Ollama[1] + Ollama WebUI[2] is a killer combination for offline/fully local LLMs. Takes all the pain out of getting LLMs going. Both projects are rapidly adding functionality including recent addition of multimodal support.

[1] https://github.com/jmorganca/ollama

[2] https://github.com/ollama-webui/ollama-webui

Yep. Edited my post to reflect as much. The MBP makes a rather nice portable package though.

Nice - that's still pretty solid.. although on a more typical 3060 or 3070 with less vram available, I probably wouldn't expect numbers quite that good.

My 14" M1 Max does around 30t/s on Mixtral Q4_K_M.

Fair enough. I did put 'ruining' in quotes for a reason - I haven't compared output between Q3 and Q4_K_M that I use, but you do generally sacrifice output quality at higher quantization levels.

And you're right, you can run it on a multi-GPU setup if you're so inclined.

Same way you'd bounce an idea or situation off a knowledgable colleague when you're stuck. Concise summary of the problem, the desired outcome, and finish the prompt with something to the tune of '[what are/walk me through] some ways I could [do/get around/prevent] this'

Mixtral 8x7B. Short of that (high ram requirements), I've also found Mistral v0.2 to be pretty solid for a 7b model.

YMMV though - it's going to depend on your use cases.

I've had excellent results with Mixtral too - it's genuinely impressive. Only problem is that it's a relatively big model that's difficult to run with full GPU inference on consumer hardware (vs the 7b/13b models people typically use).

So far, the main consumer platform capable of running it without 'ruining' the quality of its output (with high levels of quantization) is the newer Apple Silicon Macs with unified memory - generally >=48GB. It can apparently be done on 32 or 36GB, but there's not much headroom.

Edit: As coder543 points out, yes - you can run it without more lossy levels of quantization on multi-GPU setups providing those have enough combined vram.

Or try using a local LLM as a rubber ducky - that's been working pretty well for me lately.

Same here. Official Mixtral & Mistral (the new v0.2) seem to have the best & most dependable output vs basically all their predecessors/derivatives of equal or smaller parameter sizes. Mixtral is on a whole other level with its baked-in 'chain of thought' reasoning.

The fine tunes (e.g. dolphin-mixtral, dolphin2.2-mistral, etc) may be good at coding or whatever else they specialize in, but they seem to sacrifice in other areas that result in 'hallucinating' where Mixtral/Mistral wouldn't - perhaps as a result of 'score chasing' on benchmarks like has been mentioned already.

Depends on your own specific needs & use cases whether this is important or not.

How much RAM do you have? Mixtral is a beast and the non quantized model wants 40GB+ of memory.

I run 7b/13b models pretty gracefully on a 16GB M1 Pro, but it does leave me wanting a little more headroom for other things like Docker and browser eating multiple gigabytes of ram themselves.

Maybe keep an eye out for M1 / M2 deals with high ram config? I've seen 64GB MBPs lately for <$2300 (slickdeals.net)

Yeah, it's incredibly invasive. US desperately needs privacy laws governing these mass violations of privacy.

Seems like every new update, Edge gets more enabled-by-default 'telemetry' options. Here's one of the latest:

>Include related matches in Find on page

>When enabled, your Find on page search and the webpage contents will be sent to Microsoft to help find better results, including synonyms, alternate spellings, and answers to questions

Perhaps a better title for this article would be: "Business leaders are bringing a whole new vibe to the workplace: anxiety"

With the popularity of 'hot desking', overbearing micromanagement, unrealistic expectations, and continuous surveillance/activity monitoring.. who can blame people for being stressed / anxious at work?

I don't even want to think about how much time I've wasted mulling over release names.

Been using Rectangle pretty much as long as I've had my 14" M1 Pro - excellent FOSS that holds up extremely well against paid alternatives! Little to no customization necessary for a very well-executed window management experience. I did change one or two of the shortcuts to be one-handed (namely Restore window, to CTRL+OPT+R), but overall the shortcuts seem really well thought out.

I might have to grab the Pro version because they really deserve the support.

I like it so much that I wish I had the same muscle-memory for similar shortcuts on Windows (I believe PowerToys FancyZones can be set up to snap a window to a given zone or zones with hotkeys, but it's not something I've dug into yet).

Having come from Notion also, I LOVE Obsidian for its non-proprietary markdown file structure. Incredible powerful plugins, too.

FWIW, it's not really what you're seeking... but there is a plugin that allows you to invoke an LLM from within Obsidian (via Ollama): https://github.com/hinterdupfinger/obsidian-ollama

In short, it allows you to set up prompts to act on selected text directly within a file, e.g. 'Summarize this selection as a markdown formatted list of key points', 'Write a PRD', 'Translate to [Language]', 'Run this as a prompt', etc.

Despite its 'anti-fingerprint' anodization, the black still shows smudges more than silver/space gray, fwiw.. do you really need to upgrade a ~2yr old laptop??

>How does a three-year-old CPU beat the latest Apple Silicon chip in terms of performance in many popular DAWs?

Why do people keep saying the M1 Pro is 3 years old? It came out in October 2021 - it's barely over 2...

For GPT4? Sure..

For small LLMs like Llama2 7B/13B and its derivatives? They can be run quite gracefully on Apple Silicon Macs & similarly capable PC hardware.

That's not a big problem with the training/fine-tuning you would do when creating specialized 'local' LLM agents.

> My dad is restoring a 1969 MG Midget. The right turn signal stopped working. Using nothing more than a voltmeter, I found a disconnected wire and a short to the frame. I replaced the entire length of wire that was failing with $3 worth of wire, solder, and heat shrink tubing.

I want to see EVs/plug-in hybrids with similar levels of simplicity wherever possible. Virtually all vehicles in the US nowadays are completely overloaded with unnecessary sensors/electronics that are ripe for failure.

There's an easier way to get LLMs running with much less hassle if you're on MacOS: Ollama - https://ollama.ai/

I've used the 80% max charge setting on a 15 pro since the first week or so. FWIW, it seems Apple has it charge to 100% occasionally (about once a month) even with the setting active.

I wish you could have iOS artificially adjust/map the visible battery level reading to your 0-80% limit (as 0-100%) - starting at 80% gives undue battery anxiety... but the other reason I use the 80% limit is that it still lasts DAYS and I don't have to think about charging it every day. That and battery longevity is nice on a phone that may go on for 5+ years.

Single use 'Hail Mary' style rockets

Thanks for the link. I've got a Legacy that's had a 'weak' battery for a while now and it seems like it may coincide with this 3g sunsetting..