Comments on: Anthropic Fires Off Performance And Price Salvos In AI War https://www.nextplatform.com/2024/03/05/anthropic-fires-off-performance-and-price-salvos-in-ai-war/ In-depth coverage of high-end computing at large enterprises, supercomputing centers, hyperscale data centers, and public clouds. Tue, 12 Mar 2024 18:07:36 +0000 hourly 1 https://wordpress.org/?v=6.7.1 By: Eric Olson https://www.nextplatform.com/2024/03/05/anthropic-fires-off-performance-and-price-salvos-in-ai-war/#comment-221217 Wed, 06 Mar 2024 13:23:53 +0000 https://www.nextplatform.com/?p=143751#comment-221217 When a human says their feelings are not hurt, they wouldn’t be talking about feelings at all unless they were hurting.

I find it interesting Anthropic’s intelligence versus cost curve is concave up, that is, increasingly increasing as more money is spent. Of course not having a scale on one axis means the curve could have been any shape desired. Even so, the popular perception seems to be, rather than a diminishing rate of return, that there is an increasing rate of return as neural networks increase in size.

If really true, that’s an easy way to sell lots of hardware.

]]>
By: HuMo https://www.nextplatform.com/2024/03/05/anthropic-fires-off-performance-and-price-salvos-in-ai-war/#comment-221214 Wed, 06 Mar 2024 12:34:16 +0000 https://www.nextplatform.com/?p=143751#comment-221214 In reply to Timothy Prickett Morgan.

I applaud this subtle testing of emerging LLM controversies, by the TNP Studio-Garage AI Labs (TNP-SGAIL), beyond the scientific quest for etymological clarification on whether “the deliciously sensible French-born Roman emperor, who ruled between Caligula and Nero”, or, Claude Shannon (as reason, logic, and good sense “may” have it, … maybe!), is the namesake for Anthropic’s LLM, and into the recently suggested potential for metacognitive and self-awareness abilities of Claude Opus. That potential was suggested by Alex Albert on Monday ( https://twitter.com/alexalbert__/status/1764722513014329620 ) after an all-dressed pizza in a barn-full of straw bales test (or needle in a haystack).

In my current understanding then, the TNP-SGAIL’s expert Voight-Kampff testing unequivocally demonstrates, both conclusively and convincingly, that the AI replicant under study is “Not yet” an android teenager dreaming of electric sheep. This thankfully saves us from having to hire Blade Runners to administer the related pocky-clypse (or some such), at this time! d^8

]]>
By: Timothy Prickett Morgan https://www.nextplatform.com/2024/03/05/anthropic-fires-off-performance-and-price-salvos-in-ai-war/#comment-221189 Wed, 06 Mar 2024 02:22:05 +0000 https://www.nextplatform.com/?p=143751#comment-221189 In reply to Hubert.

I asked Claude 3 Sonnet if it knew what it was named for, and it said it didn’t, and I then told it that it was probably Claude Shannon, and its creators could have shown it courtesy and respect by telling Claude its own origin story. To which, it said it could see how I was right but it didn’t have any feelings to be hurt.

Not yet, anyway. HA!

]]>
By: Hubert https://www.nextplatform.com/2024/03/05/anthropic-fires-off-performance-and-price-salvos-in-ai-war/#comment-221187 Wed, 06 Mar 2024 01:44:44 +0000 https://www.nextplatform.com/?p=143751#comment-221187 A great article for Super Tuesday (the French LCI (24h TV News) live-streams ABC NewsLive as we speak (2am here, 8pm in US))!. This AI-model pricing analysis suggests (to me) that the more accurate LLMs (Opus, GPT-4) are currently similar to handcrafted luxury goods when compared to the less accurate “mass-consumption” ones (Haiku, Gemini Pro). As the tech matures (in the future) we’ll hopefully reach a point where costs increase slower than performance, such that the higher-end models end-up costing less on a price/perf basis (as with commodity cheeseburgers, at one for $1.39, and two for $2.19, …). Not a big shock here though, as France’s LVMH (Louis Vuitton, Moët, Hennessy) is the world leader of luxury goods (eh-eh-eh!)!

I quite like that Claude Shannon and mouse (Theseus, the minotaur’s maze solver) picture! A world-first example of machine learning from 1950 ( https://www.technologyreview.com/2018/12/19/138508/mighty-mouse/ ).

]]>