Conversation

Notices

1. Embed this notice
   God Emperor of Mastodon (mms@mastodon.bsd.cafe)'s status on Thursday, 04-Sep-2025 02:33:41 JST God Emperor of Mastodon

   https://futurism.com/ai-far-away-profit-experts-warn

   "The AI Industry Is Still Light-Years From Making a Profit, Experts Warm"

   ♫ It's getting cold in here ♫

   • Embed this notice
     I am Water (slicerdicer@friedcheese.us)'s status on Thursday, 04-Sep-2025 02:33:39 JST I am Water

     in reply to
     @mms It’s gonna be a face plant from hell.
     
     No pizza, no cookies, just dual wielding cigarettes as the value vaporizes.
     Doughnut Lollipop 【記録係】:blobfoxgooglymlem: likes this.
   • Embed this notice
     I am Water (slicerdicer@friedcheese.us)'s status on Thursday, 04-Sep-2025 02:37:09 JST I am Water

     in reply to

     • feld
     @feld @mms That means being exciting and having bait.
     Doughnut Lollipop 【記録係】:blobfoxgooglymlem: likes this.
   • Embed this notice
     feld (feld@friedcheese.us)'s status on Thursday, 04-Sep-2025 02:37:10 JST feld

     in reply to

     • I am Water
     @SlicerDicer @mms whoever can keep the investors for long enough will win, essentially.
   • Embed this notice
     feld (feld@friedcheese.us)'s status on Thursday, 04-Sep-2025 02:37:11 JST feld

     in reply to

     • I am Water
     @SlicerDicer @mms Amazon survived this same scenario
     
     > For years, Amazon posted losses, and it wasn't until 2003 – almost a full decade later – that Amazon finally turned a full-year profit.
   • Embed this notice
     I am Water (slicerdicer@friedcheese.us)'s status on Thursday, 04-Sep-2025 02:37:11 JST I am Water

     in reply to

     • feld
     @feld @mms Market is less forgiving now.
     
     We have less attention span.
     Doughnut Lollipop 【記録係】:blobfoxgooglymlem: likes this.
   • Embed this notice
     Doughnut Lollipop 【記録係】:blobfoxgooglymlem: (tk@bbs.kawa-kun.com)'s status on Thursday, 04-Sep-2025 02:40:19 JST Doughnut Lollipop 【記録係】:blobfoxgooglymlem:

     in reply to

     • I am Water

     @SlicerDicer @mms As long as the tech industry exists, this will continue to be relevant. :blobfoxgooglytrash:

   • Embed this notice
     Doughnut Lollipop 【記録係】:blobfoxgooglymlem: (tk@bbs.kawa-kun.com)'s status on Thursday, 04-Sep-2025 02:40:32 JST Doughnut Lollipop 【記録係】:blobfoxgooglymlem:

     in reply to

     • feld
     • I am Water
     @SlicerDicer @feld @mms Snake oil. :blobfoxgooglytrash:
   • Embed this notice
     I am Water (slicerdicer@friedcheese.us)'s status on Thursday, 04-Sep-2025 02:40:53 JST I am Water

     in reply to

     • feld
     @mms @feld The entire Seattle area was basically in a depression too between dot com bust and 9/11 wasting Boeing. Costs were extremely low at the time.
     Doughnut Lollipop 【記録係】:blobfoxgooglymlem: likes this.
   • Embed this notice
     God Emperor of Mastodon (mms@mastodon.bsd.cafe)'s status on Thursday, 04-Sep-2025 02:40:54 JST God Emperor of Mastodon

     in reply to

     • feld
     • I am Water

     @feld @SlicerDicer Amazon wasn’t solely responsible for all stock growth nor required recreation of us electric grid. They were small potatoes

   • Embed this notice
     Doughnut Lollipop 【記録係】:blobfoxgooglymlem: (tk@bbs.kawa-kun.com)'s status on Thursday, 04-Sep-2025 02:55:30 JST Doughnut Lollipop 【記録係】:blobfoxgooglymlem:

     in reply to

     • feld
     • I am Water

     @feld @SlicerDicer @mms How many billions of queries is it handling, though? The total energy consumption per unit time is what really matters here.

   • Embed this notice
     feld (feld@friedcheese.us)'s status on Thursday, 04-Sep-2025 02:55:31 JST feld

     in reply to

     • feld
     • I am Water
     @mms @SlicerDicer this also does not even take into consideration what Google is already seeing. Their Gemma model is reportedly only consuming "0.24 watt-hours (Wh) of energy, emits 0.03 grams of carbon dioxide equivalent (gCO2e), and consumes 0.26 milliliters (or about five drops) of water" per the average query.
     
     That energy usage is the same as a microwave for one second.
     
     
     https://cloud.google.com/blog/products/infrastructure/measuring-the-environmental-impact-of-ai-inference/
   • Embed this notice
     feld (feld@friedcheese.us)'s status on Thursday, 04-Sep-2025 02:55:32 JST feld

     in reply to

     • I am Water
     @mms @SlicerDicer I think the power consumption concerns are way overblown. The advances happening are incredible. You will be able to get hundreds to thousands of tokens per second out of consumer hardware soon. And not require massive amounts of memory to run the models.
     
     And that doesn't even include other advances like Cerebras which has custom ASICs that drop the power consumption significantly and have 10x the performance. And they are just getting started.
     
     https://www.cerebras.ai/blog/cerebras-cs-3-vs-nvidia-b200-2024-ai-accelerators-compared
     
     The design of the transformers/engines that run the models right now are incredibly inefficient and stupid, because doing it right is "hard". But there are people doing it right and it will blow people's minds when they see what gets released in the next year or two.
   • Embed this notice
     I am Water (slicerdicer@friedcheese.us)'s status on Thursday, 04-Sep-2025 02:57:22 JST I am Water

     in reply to

     • Doughnut Lollipop 【記録係】:blobfoxgooglymlem:
     • feld
     • I am Water
     @tk @feld @mms If you have to grep an indexed code base or read files on an indexed codebase what’s the point of indexing? When you use thousands of input tokens that require qkv and mlp where are you at? This is wasteful as sin.
     Doughnut Lollipop 【記録係】:blobfoxgooglymlem: likes this.
   • Embed this notice
     I am Water (slicerdicer@friedcheese.us)'s status on Thursday, 04-Sep-2025 02:57:23 JST I am Water

     in reply to

     • Doughnut Lollipop 【記録係】:blobfoxgooglymlem:
     • feld
     @tk @feld @mms You are missing the point of efficiency. Less work done is more work done when it comes down to it. The design I’m doing is based in reality not hype.
     
     Current methods are enormously wasteful.
     Doughnut Lollipop 【記録係】:blobfoxgooglymlem: likes this.
     Doughnut Lollipop 【記録係】:blobfoxgooglymlem: repeated this.
   • Embed this notice
     I am Water (slicerdicer@friedcheese.us)'s status on Thursday, 04-Sep-2025 03:06:47 JST I am Water

     in reply to

     • Doughnut Lollipop 【記録係】:blobfoxgooglymlem:
     • feld
     • I am Water
     @tk @feld @mms I should clarify when it reads line 200-500 on file.extension every prompt. What does this do to context? What does this do for qkv and mlp?
     
     Same shit computed over and over and over again. Yet it still doesn’t seem to understand dependency chain and will generate duplicate code. Thats abomination. The work I’m doing solves this.
     Doughnut Lollipop 【記録係】:blobfoxgooglymlem: likes this.