23 Foundation Models Are Improving Across Domains With larger training datasets and more parameters, GPT-4 outperforms GPT-3.5 significantly. Increasingly, foundation models are becoming “multimodal”—supporting text, images, audio, and video—and are not only more dynamic and user friendly, but also more performant. GPT-3.5, GPT-4, and Claude 2 Results on Professional and Academic Exams GPT-3.5 GPT-4 Claude 2 GPT-4 Vision E 100% C N E 90% G I L 80% L E 70% T N e I il 60% AL nt I ce 50% C I r F e 40% I P T AR 30% 20% 10% 0% 0 al g e h y T t y s y s y s e y 2 y 2 C e e 0 g W m t n r c g c g c r 1 r 1 2 c r A i i i v r s g n b n a a o i o B u i 0 r R n x o e t t o o t t c t C s a t C t ti t s l m l m s i 2 e B i e M LS m s i i s s u a a r i E s i t o o o o ta i u g r l V E c i i i y l R a r T n H a n h n t m H AM n e AM T W S a H r t B c e h u t s n E l e t S o y o n P d c a i e fi SA E a B SA S v c s c a h l al L L c i GR t U o Ar AP e P e u C r h h r m GR n rm G AP o o Q AP o C s s o e e o AP r r W li li f S m f AP S ac AP ic E AP AP g g e * n i U M M n n d O o Un GR AP E E Co B ir AP AP AP A nv AP AP US E AP *USA Biology Olympiad, a prestigious national competition testing high school students in biology. Sources: ARK Investment Management LLC, 2024, based on data from OpenAI and Anthropic as of Jan. 9, 2024. Forecasts are inherently limited and cannot be relied upon. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security. Past performance is not indicative of future results.