Anthropic's Great AI Growth: Claude 3.7 Sonnet Taught for Tens of Year

Claude 3.7 Sonnet, Anthropic's latest artificial intelligence marvel, can be taught for "tens of millions of dollars" with fewer than 10^26 FLOPs of raw computing power!

In a fascinating revelation, Ethan Mollick, a respected Wharton professor, shared some exciting news in an X post he published on Monday. It appears he had some enlightening chitchat with the folks at Anthropic's PR team. "Anthropic made it clear to me that Sonnet 3.7 is not considered a 10^26 FLOP model and it had a price tag of a few tens of millions of dollars for training". He also spilled the beans about the future - the upcoming models' expected size gave us quite a bit to ponder on. We're talking massive, folks!

On reaching out to Anthropic for a solid confirmation about this claim, the peeps over at TechCrunch hadn't received a resounding 'yes' or 'no' when we last checked.

Claude 3.7 Sonnet's training cost of "tens of millions of dollars" is an interesting example of the revolutionary steps being taken to make advanced AI models affordable. The predecessor, Claude 3.5 also carried a similar training price tag as revealed in a candid essay by Anthropic's head honcho, Dario Amodei.

These figures come across as highly favorable especially when compared to the whopping amounts shelled out in 2023 by AI giants for their respective models. OpenAI put forth more than $100 million to raise its brainchild, GPT-4 model, as stated by OpenAI's CEO Sam Altman himself. Not to be outdone, Google shelled out nearly $200 million to train its Gemini Ultra model, as suggested by a Stanford study.

However, Amodei is on his toes anticipating that the future behemoths of AI would come with a billion-dollar price tag. Besides, let's not forget that these training costs don't take into account significant groundwork like safety testing and fundamental research. Also, as the AI industry leans towards implementing "reasoning" models willing to mull over problems for an extended duration, we can expect a surge in the running costs of these models. Stay tuned for more fascinating updates from the world of AI!

by rayyan