#392: OpenAI’s Improved ChatGPT Should Delight Both Expert And Novice Developers, & More

Last week, Founder Sam Altman led OpenAI’s first developer event[i] with news of significant model improvements and an innovative service for creating custom versions of ChatGPT.

OpenAI trained its latest model, GPT-4 Turbo, on data through April 2023 with a much larger, 128K-token context window that can ingest ~300 pages of a book in one go and can perform with much less latency. Developers can use GPT-4 Turbo at a cost ~2.75 times lower on average than the original GPT-4.

In addition, OpenAI users now can create customized assistants to engage GPT with natural language instructions, no coding required. As powerful extensions of ChatGPT’s natural language interface, custom agents can perform detailed tasks, write code, call external APIs, and integrate with private knowledge databases. Novice developers can use the simple chat interface to create custom GPTs in minutes, and popular GPT developers can pay for “GPT Stores.”

OpenAI is capitalizing on the early success of ChatGPT. Allowing anyone to create and host their unique versions of ChatGPT on the platform, OpenAI could be the reincarnation of Apple’s App Store. In other words, ChatGPT could become a universal interface and the first point of user contact, answering questions and executing tasks by communicating with user-generated GPTs.

While the performance of GPT-4 Turbo and other closed AI models is improving, open-source models are gaining ground quickly. Recently, China’s 01.AI released an open-source model, Yi-34B, that outperformed its open-source peers on several benchmarks. On MMLU, a benchmark measuring basic knowledge across 57 subjects, for example, Yi-34B scored 76.3%, surpassing the 68.9% and 70.4% of other open-source models like Meta Platforms’ LLaMA2-70B and the TII’s Falcon-180B, respectively. Transparent and free, open-source models seem to be benefiting from the contributions of diverse developer groups that are accessing and contributing to the leading foundation models.

ARK’s research corroborates that the performance gap between private and open-source models is narrowing, as shown below. While OpenAI’s GPT-4 still outperforms all other competitors, the trendlines below suggest that open-source models are gaining ground and could be only months behind the performance of more costly, private, closed-source models.

Source: ARK Investment Management LLC, 2023. Data as of November 10, 2023. This ARK analysis is based on a range of data sources, which are available upon request. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency. Past performance is not indicative of future results.

Last week, bitcoin futures traded[ii] on the Chicago Mercantile Exchange (CME) hit two major milestones. First, the open interest in bitcoin futures surpassed 100,000 BTC, hitting an all-time high. Second, the CME surpassed Binance as the exchange with the highest open interest in bitcoin futures: CME’s 113,000 BTC at 27.7% of the market compared to Binance’s 104,000 at 26.7% of the market.

In our view, the increased activity in CME contracts suggests that institutions have entered the market, as the minimum per contract is 5 bitcoin, or ~$175,000—a high barrier to entry. Adding to the momentum, this week the bitcoin price surpassed $35,000 for the first time since May 2022—when Terra, LUNA, and UST collapsed.[iii] Other catalysts for continued bitcoin price appreciation include the potential SEC approval of a spot bitcoin ETF and the halving of bitcoin’s issuance, historically a bullish event, in April 2024.

[i] OpenAI DevDay. 2023. “Opening Keynote.” YouTube.
[ii] @WClementeIII. 2023. “Wow, the real flippening that no one is talking about.” X. 
[iii] Sandor, K. and E. Genç,. 2022. “The Fall of Terra: A Timeline of the Meteoric Rise and Crash of UST and LUNA.” CoinDesk.