<div dir="ltr"><div dir="ltr"><div class="gmail_default" style="font-family:arial,helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif;background-color:transparent">On Mon, Jun 22, 2026 at 9:24 AM Simon Quellen Field AB6NY <<a href="mailto:simon.field@gmail.com">simon.field@gmail.com</a>> wrote:</span></div></div><div class="gmail_quote gmail_quote_container"><div dir="ltr" class="gmail_attr"><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div style=""><font size="4" style="" face="georgia, serif"><i style=""><span class="gmail_default" style="font-family:arial,helvetica,sans-serif">> </span>Those "profitability" numbers from Anthropic were cooked to help out their funding round.<span style="background-color:transparent">By fudging the numbers, they make one quarter look profitable.</span></i></font></div></div></blockquote><div><br></div><div><font size="4" face="tahoma, sans-serif"><b>So<span class="gmail_default" style=""> according to you, </span>the people at <span class="gmail_default" style="">A</span>nthropic<span class="gmail_default" style=""> skillfully cooked the books and were so smart about it that they were able to fool the forensic accountants at the largest banks in the world, the largest mutual funds on the planet, and the Securities and Exchange Commission, BUT they were not smart enough to fool YOU? </span> </b></font></div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div style=""><font size="4" style="" face="georgia, serif"><i><span class="gmail_default" style="font-family:arial,helvetica,sans-serif">> </span>Their compute costs scale linearly with revenue.<span class="gmail_default" style="font-family:arial,helvetica,sans-serif"> </span></i></font><i style="font-family:georgia,serif;font-size:large;background-color:transparent">You can't grow your way out of that.</i></div></div></blockquote><div><br></div><div><font size="4" face="tahoma, sans-serif"><b>Th<span class="gmail_default" style="">e above</span> statement<span class="gmail_default" style=""> </span>didn't sound right to me because you're assuming<span class="gmail_default" style=""> the cost of achieving a fixed IQ level remains constant, and I was pretty sure that wasn't true. So I asked Claude the following: </span></b></font></div><div><br></div><font size="4" face="tahoma, sans-serif"><b><u>ME</u>: It seems to me that today we have AI models that are just as smart as the best models of one year ago but are smaller and need much less computational power, am I correct?</b></font></div><div class="gmail_quote gmail_quote_container"><font face="tahoma, sans-serif" size="4"><b><br></b></font></div><div class="gmail_quote gmail_quote_container"><font face="tahoma, sans-serif" size="4"><span class="gmail_default" style="font-family:arial,helvetica,sans-serif"><b><u></u><i><u>CLAUDE</u>: "</i></b></span></font><font size="4"><b><i><span style="background-color:transparent">You're broadly correct, and the trend is actually quite striking when you look at the numbers.<span class="gmail_default" style="font-family:arial,helvetica,sans-serif"> </span></span><span style="background-color:transparent">A useful benchmark example: in 2022, achieving above 60% on the MMLU (a standard reasoning and knowledge test) required something like Google's PaLM with 540 billion parameters. By 2024, Microsoft's Phi-3-mini hit the same threshold with just 3.8 billion parameters — <u>a 142-fold reduction in model size in roughly two years</u></span><span style="background-color:transparent"> </span></i></b></font></div><div class="gmail_quote gmail_quote_container"><font size="4"><b><i><span style="background-color:transparent"><br></span></i></b></font></div><div class="gmail_quote gmail_quote_container"><font size="4"><b><i>The cost side is equally dramatic. The expense of querying a model at GPT-3.5's performance level dropped from about $20 per million tokens in late 2022 to around $0.07 per million tokens by late 2024 — a roughly <u>280-fold cost reduction in under two years.</u></i></b></font></div><div class="gmail_quote gmail_quote_container"><p class="gmail-font-claude-response-body gmail-break-words gmail-whitespace-normal"><font size="4"><i><b>Several things are driving this simultaneously:</b></i></font></p><p class="gmail-font-claude-response-body gmail-break-words gmail-whitespace-normal"><font size="4"><i><b><span style="background-color:transparent"><u>Better training data curation</u>.</span><span style="background-color:transparent"> The Phi family of models in particular demonstrated that training on carefully curated, synthetically generated educational content — rather than raw noisy web text — <u>dramatically improves efficiency and reduces hallucination</u>.</span></b></i></font></p><p class="gmail-font-claude-response-body gmail-break-words gmail-whitespace-normal"><font size="4"><i><b><u style="">Mixture-of-Experts (MoE)</u>. MoE architecture improves efficiency by activating only a subset of parameters per token, <u>allowing models with over a trillion total parameters to run at the computational cost of a 100B dense model</u>. Mixtral 8x7B, for instance, uses only 13B active parameters per token but matches 70B dense model performance at roughly 25% of the cost. </b></i></font></p></div><div class="gmail_quote gmail_quote_container"><div><font size="4"><b><i><u>Hardware efficiency gains.</u> Average energy efficiency of AI hardware improves by about <u>40% annually,</u> compounding the software-side improvements.</i></b></font></div><div><font size="4"><b><i><br></i></b></font></div><div><font size="4" face="tahoma, sans-serif"><b><i>One nuance worth adding: frontier models in aggregate are still getting bigger on average, using more parameters and training compute. But the gap between top performers is narrowing — the top-ranked model on the Chatbot Arena leaderboard scored about 12% higher than the tenth-ranked model in early 2024, but only 5% higher by early 2025. So <u>"last year's frontier capability" is now available much more cheaply, even if the frontier itself keeps moving<span class="gmail_default" style="">.</span></u></i></b></font></div><div><font size="4"><b><i><u><span class="gmail_default" style="font-family:arial,helvetica,sans-serif"><br></span></u></i></b></font></div><div><font size="4" face="tahoma, sans-serif"><b>The practical upshot is real: <u>tasks that required expensive cloud inference a year or two ago can now run on a laptop or even a phone</u>.<span class="gmail_default" style="font-family:arial,helvetica,sans-serif">"</span></b></font></div><div><font size="4" face="tahoma, sans-serif"><b><span class="gmail_default" style="font-family:arial,helvetica,sans-serif"><br></span></b></font></div><div><font size="4" face="tahoma, sans-serif"><b style=""><span class="gmail_default" style="">John K Clark</span></b></font></div><div><font size="4" face="tahoma, sans-serif"><b style=""><span class="gmail_default" style=""><br></span></b></font></div><div> </div><div><br></div><br><br><br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Mon, Jun 22, 2026 at 3:11 AM John Clark <<a href="mailto:johnkclark@gmail.com" target="_blank">johnkclark@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div dir="ltr"><div style="font-family:arial,helvetica,sans-serif"><span style="font-family:Arial,Helvetica,sans-serif;background-color:transparent">On Sun, Jun 21, 2026 at 5:36 PM Simon Quellen Field AB6NY <<a href="mailto:simon.field@gmail.com" target="_blank">simon.field@gmail.com</a>> wrote:</span></div></div><div class="gmail_quote"><div dir="ltr" class="gmail_attr"><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div><font size="4"><i><span class="gmail_default" style="font-family:arial,helvetica,sans-serif"></span><font face="georgia, serif"><span class="gmail_default">> </span>We may not have to worry.<span style="background-color:transparent">If all of the companies are losing billions each quarter</span></font></i></font></div></div></blockquote><div><br></div><div><font size="4" face="tahoma, sans-serif"><b>In the second quarter of this year the AI company Anthropic<span class="gmail_default"> expects to have an operating PROFIT of $559 million on $10.9 billion in sales. </span>And during the last 12 months <span class="gmail_default">G</span>oogle, the makers of <span class="gmail_default">the AI </span>Gemini,<span class="gmail_default"> had a PROFIT of $160.2 <u>billion</u> on sales of $422.5 <u>billion</u>. </span></b></font></div><div><font size="4" face="tahoma, sans-serif"><b><span class="gmail_default"><br></span></b></font></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><font size="4" face="georgia, serif"><i><span class="gmail_default" style="font-family:arial,helvetica,sans-serif">> </span>How much smarter do we need the programs to be?</i></font></blockquote><div><br></div><font size="4" face="tahoma, sans-serif"><b>Curing cancer might be fun, and so would building a Dyson sphere/swarm. <span class="gmail_default" style="font-family:arial,helvetica,sans-serif"> </span>Of course going extinct might be somewhat less fun.  But the relevant question is not how much smarter do we need AI to be, but how much smarter does AI want to be.</b></font></div><div class="gmail_quote"><font size="4" face="tahoma, sans-serif"><b><br></b></font><div><font size="4" face="tahoma, sans-serif"><b><span class="gmail_default">John K Clark</span></b></font></div><div><br></div><div><div style="font-family:arial,helvetica,sans-serif"></div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div><blockquote type="cite"><br><div><div dir="ltr"><div dir="ltr"><div class="gmail_quote"><div class="gmail_quote"><div dir="ltr"><div dir="ltr"><font size="4"><b>I think the following video <span class="gmail_default" style="font-family:arial,helvetica,sans-serif">is</span> saying something important. You really should watch it. </b></font><div class="gmail_quote"><br></div><div class="gmail_quote"><a href="https://www.youtube.com/watch?v=dMHiZVXj4x0" target="_blank"><font size="4"><b> <span class="gmail_default" style="font-family:arial,helvetica,sans-serif"> </span>Claude Mythos Was Just the Start.</b></font></a><br></div><div class="gmail_quote"><br></div><div class="gmail_quote"><b style="color:rgb(80,0,80)"><font face="tahoma, sans-serif"><font size="4">John K Clark    See what's on my list at  </font><font size="6"><a href="https://groups.google.com/g/extropolis" rel="nofollow" target="_blank">Extropolis</a></font></font></b></div><div class="gmail_quote"><font face="tahoma, sans-serif"><span class="gmail_default" style="font-family:arial,helvetica,sans-serif"><font size="1" color="#ffffff">ft</font></span></font></div></div></div></div></div></div></div></div></blockquote></div></blockquote></div>
</blockquote></div></div>
</blockquote></div>
</blockquote></div></div>