Jump to content

Goredho

Certifiably Surly
  • Posts

    8247
  • Joined

  • Days Won

    1

Everything posted by Goredho

  1. RFK Jr looks like Buffalo Bill from Silence of the Lambs if he targeted octogenarians from Florida to make his skin suit out of.
  2. No One: Trump Voters: SESAME STREET GROOMED MY KIDS TO BE FURRIES!!!
  3. I could play air guitar with those.
  4. All these budget brands have their place, but I have owned 3 Harley Bentons, and they a project to get playing/sounding great. Better than not having a guitar if that's what your budget is. But if you want it to play and sound great, you'll have some work to do and/or money to spend. You can get a great guitar that isn't a project beyond a setup in the $500-$800 range. If you start with a $200 budget guitar and need to put $300 worth of electronics/luthier labor into it to play and sound like a $1000 guitar, why not buy a used Reverend or PRS SE for roughly the same amount and just set it up and play? I get some people love the tinkerers aspect and the sweat equity they put into making a cheap guitar great, but I decided I'd rather just play.
  5. https://simonwillison.net/2025/Jan/29/on-deepseek-and-export-controls/#atom-everything On DeepSeek and Export Controls. Anthropic CEO (and previously GPT-2/GPT-3 development lead at OpenAI) Dario Amodei's essay about DeepSeek includes a lot of interesting background on the last few years of AI development. Dario was one of the authors on the original scaling laws paper back in 2020, and he talks at length about updated ideas around scaling up training: The field is constantly coming up with ideas, large and small, that make things more effective or efficient: it could be an improvement to the architecture of the model (a tweak to the basic Transformer architecture that all of today's models use) or simply a way of running the model more efficiently on the underlying hardware. New generations of hardware also have the same effect. What this typically does is shift the curve: if the innovation is a 2x "compute multiplier" (CM), then it allows you to get 40% on a coding task for $5M instead of $10M; or 60% for $50M instead of $100M, etc. He argues that DeepSeek v3, while impressive, represented an expected evolution of models based on current scaling laws. [...] even if you take DeepSeek's training cost at face value, they are on-trend at best and probably not even that. For example this is less steep than the original GPT-4 to Claude 3.5 Sonnet inference price differential (10x), and 3.5 Sonnet is a better model than GPT-4. All of this is to say that DeepSeek-V3 is not a unique breakthrough or something that fundamentally changes the economics of LLM's; it's an expected point on an ongoing cost reduction curve. What's different this time is that the company that was first to demonstrate the expected cost reductions was Chinese. Dario includes details about Claude 3.5 Sonnet that I've not seen shared anywhere before: Claude 3.5 Sonnet cost "a few $10M's to train" 3.5 Sonnet "was not trained in any way that involved a larger or more expensive model (contrary to some rumors)" - I've seen those rumors, they involved Sonnet being a distilled version of a larger, unreleased 3.5 Opus. Sonnet's training was conducted "9-12 months ago" - that would be roughly between January and April 2024. If you ask Sonnet about its training cut-off it tells you "April 2024" - that's surprising, because presumably the cut-off should be at the start of that training period? The general message here is that the advances in DeepSeek v3 fit the general trend of how we would expect modern models to improve, including that notable drop in training price. Dario is less impressed by DeepSeek R1, calling it "much less interesting from an innovation or engineering perspective than V3". I enjoyed this footnote: I suspect one of the principal reasons R1 gathered so much attention is that it was the first model to show the user the chain-of-thought reasoning that the model exhibits (OpenAI's o1 only shows the final answer). DeepSeek showed that users find this interesting. To be clear this is a user interface choice and is not related to the model itself. The rest of the piece argues for continued export controls on chips to China, on the basis that if future AI unlocks "extremely rapid advances in science and technology" the US needs to get their first, due to his concerns about "military applications of the technology".
  6. Interesting... https://www.theguardian.com/technology/2025/jan/28/we-tried-out-deepseek-it-works-well-until-we-asked-it-about-tiananmen-square-and-taiwan
  7. I am sorry for your loss, Brisket. I lost my dad in 2005 and my mom in 2022. It's very, very weird being parentless. To not have those giants in your corner any more, and to think it is now your responsibility to somehow fill their void in the world. I can't say anything to make it easier. I would just urge you to not rush whatever grieving process you need to go through. And I would think about a fishing trip for next Father's Day, someplace in solitude, so you can be alone with your dad.
  8. Soaking in water helps to bring out the starch so they fry up crisper.
  9. Yep, I very well could be naive that this will change anything within our company or outside of it. But I hope something does, or the "broken eggs" will be worse than a few hundred billion market cap across a handful of companies. Whether or not that could be correctly characterized as broken eggs or catastrophic failures will largely depend on where one sits on a spectrum of sociopathy.
  10. Listen to @Captainant, @raw dog. He does this for a living, every day. We're in the peer review stage. If this is just Chinese propaganda, it will be a blip after being exposed as such in short order. If its not, its going to upend the game table for AI as alluded to in that post and as seen in the market yesterday. I don't think the Chinese would release DeepSeek as (sorta) open source if it would not pass scrutiny, but we will see. The problem is not looking at how to make differentiating products and intellectual property with AI. Its being pressured to provide "AI YESTERDAY" so that everyone is doing piss-poor work and slapping an "AI" sticker on it to check the AI box as quickly as possible to placate people who have no idea what they are pushing for or the risks being taken. We are one three mile island event at the hands of an AI managed system from "AI" becoming a dirty word. We're creating a context where that is more likely to happen, though this event will just cost hundreds of billions vs any human lives. I am hoping this event instills some sobriety in AI-drunk executives/directors/investors.
  11. Yep, every BOD meeting I am in for our company, the first question is, "What are you doing to bring AI to bear for our business?" No consideration for how AI might be applicable, just "We need AI, we need it yesterday." In the long run this will all be good for the industry, and some of that good would be dampen the pressure to wantonly adopt AI, any AI.
  12. Well, it seems like the Chinese went into the problem from a different perspective and found an elegant, simplified solution to the compute gluttony associated with training models. This happens a lot, but not at this scale of investment and publicity. Something foundational was missed by all the engineers for all the big AI players in the US who are used to operating with unlimited compute and capital. It really calls into question their leadership in the space. Its an oversimplified exaggeration, but billions of investment dollars have been sunk into what may wind up being a dead branch of the evolutionary tree of AI and at the urging of American technologists. How are investors going to react to that? There are going to be some tense board of director meetings coming up. Good article about the market implications. https://daringfireball.net/linked/2025/01/27/nvidia-deepseek-haircut To make a long story short, it’s been broadly assumed that leading-edge model training and inference requires the highest of high-end hardware — which hardware, especially for training, pretty much exclusively comes from Nvidia. The assumption was that any company trying to compete in this fiercely competitive field — a field with massive implications for industry, culture, and national security — must get in line to buy (or rent) enormous amounts of computational power exclusively available on the very best systems from Nvidia. That hardware isn’t available (at least legally) to Chinese companies, because of the Biden administration’s export bans. DeepSeek engineers found and implemented multiple massive efficiency improvements that allowed the company to train its latest models at far lower prices using far-less-capable hardware, and perform inference under heretofore unprecedented memory constraints. DeepSeek’s achievements render obsolete several prior planks of conventional wisdom regarding the state of AI. My link a few sentences prior in this paragraph is to a WSJ report from 14 days ago that now feels like it dates from at least 14 months ago.
  13. Those states should deport him back to Texas. LA MIGRA! VAMANOS!
  14. I mean, DDOSing government services with bullshit was a valid tactic before the election. Why stop now?
  15. Seen posted in a FB neighborhood page. ICEing will be the new SWATing. Hey Leopards! Look over there!!
  16. I read that in Al Swearengen‘s voice.
  17. Do it, Kylo Ren.
  18. Dude, have you seen my gear buying habits? I’m as patient as a teenagers first orgasm. Looking forward to playing it in 2 weeks!
  19. Cool. I would look at the Positive Grid Spark, too. It’s a pretty great product, but is pretty strictly in the “practice amp” space. I couldn’t imagine using it at a gig, but it’s primarily controlled via phone app and they’ve done a good job with app interface, etc… if you could have something that combined the small gig versatility of the Laney I linked with the modeling/control/app of the Spark that had a docking station for your phone, I’d be very interested in the product. PS I expect a prototype to beta test 😜
  20. Gotcha. So that Laney amp would be a direct market competitor for what you are investigating. I would say it works great, but for guitars, you need that other thing that will emulate the guitar amp and cab, then the Laney or any such product just amplifies that with little to no colour. A lot of people just use a PA. I would look at the new Fender FR-10 as a competitor, but it’s just kind of marketing itself as FRFR for guitar amp modeling. I use modelers live all the time. I don’t use via laptop, phone or tablet. I go for UAFX pedals and the Tonex One pedal. I go for pedal-based hardware for a couple reasons. First, a tablet feels fragile. Knock it off a stand, there goes your gig. Second is more logistic and how it incorporates it into a signal chain with pedals and all that. I don’t want to connect tablet to interface, have stuff running from my feet to the interface on a stand, back down to crap at my feet, etc…. With it all being pedalboard based, it can all stay wired up. I pull out the pedalboard, plug guitar into the first pedal in the chain, plug the last pedal into an FRFR speaker like the Laney and then go. I’m lazy like that. A new product that could mitigate those issues would be interesting. I would look into the speaker cabinet providing a docking station for your phone so the phone physically becomes one with the speaker/cabinet, then have a traditional amp fx loop in the back so you can connect it with a pedalboard like you would a normal tube amp. The phone would provide the DSP brain with whatever software you provide. You also might want to provide an impulse response loader in the speaker cabinet to provide speaker sounds so that doesn’t need to draw from the phone.
  21. Traveling today, but would look at Laney amps for role number one. I have this one, and it’s great. 12” plus horn, 5 channels, fx loop, built in EQ and delay, line out to go to a PA if you need more reinforcement. I’ve run guitars through it (acoustic w/preamp and electric w/modeler), bass (w/modeler/preamp), keyboard (straight), mics (XLR) and even iPhone for backing tracks with RCA inputs. I’ve had a trio use this as the only amplification in a small bar/coffee shop setting. It’s a great Swiss Army knife that’s basically a 5 channel PA, amp and full range speaker built into a 1x12 combo enclosure and is designed exactly for what you are describing. Will see about answering B in a few. https://www.sweetwater.com/store/detail/AH150--laney-ah150-150w-12-inch-multi-input-combo-amp?mrkgadid=1000000&mrkgcl=28&mrkgen=gdsa&mrkgbflag=0&mrkgcat=studio&recording&acctid=21700000001645388&dskeywordid=39700065017603469&lid=39700065017603469&ds_s_kwgid=58700007231540659&device=m&network=g&matchtype=&adpos=largenumber&locationid=9029128&creative=706902477314&targetid=aud-439454427786:dsa-1385849839650&campaignid=13710079600&awsearchcpc=1&gad_source=1&gclid=Cj0KCQiA4rK8BhD7ARIsAFe5LXJzyJO4stbVUOj_PMKGtIQpbxLH60dGjPDGxcg-l892ASy68exL2F0aAjHZEALw_wcB&gclsrc=aw.ds
  22. This is true. Our local high school's music teacher is a very nice lady well versed in theory and playing the piano. But I've been volunteering to help her because more than half the people she has in music class wants to play guitar. So her and I collaborate on how to translate what she has to teach to the guitar for the students that are focusing on it. Was pretty cool last year, they did a really nice concert culminating with a pretty damn good rendition of Journey's "Seperate Ways" as if performed by Lynyrd Skynyrd and 2134091234 guitarists. I have no idea how they settled on that song or the rest of their setlist. Its a small town that time forgot, apparently.
  23. This is great advice. I had a piano in the house from infancy, and at one time I was a pretty decent piano player. My older brother got a guitar in high school, he had a friend who was a drummer. They needed a bassist, so I got drafted. Anyway, through middle school on high school, I had a piano, gutiar and bass in the house, and that as much as anything has been a foundation of a lifelong love affair with music (though I had about a 10 year hiatus while kids were young).
×
×
  • Create New...