OpenAI Developer Conference: Customizable AI Sparks Excitement and Concern

Ben Lorica

2 years ago

After spending several hours reading discussion threads, social media posts, and technology press articles, I observed a range of reactions to OpenAI’s announcement that users can create custom versions of ChatGPT. These reactions reflect both optimism about the potential benefits of this new functionality and concerns about increased platform control.

Many welcome the new functionality as enhancing accessibility and flexibility. Potential benefits include custom conversational personas and domain-specific expertise, enabling more nuanced and specialized applications tailored to individual needs. The substantially expanded 128K context size of GPT-4 Turbo prompted particular enthusiasm, with users noting its potential for more sophisticated conversations and reasoning. Appreciation was also expressed for the reduced pricing and new modalities like computer vision, suggesting these developments make advanced AI more financially and technically attainable for a broader range of users.

However, alongside this optimism were apprehensions about OpenAI potentially exerting too much influence over the AI landscape. Fears are surfacing that the custom GPTs may allow OpenAI to capture more value from the ecosystem by incorporating promising third-party innovations into its own platform once they gain traction. The proprietary nature of the platform evokes concerns about vendor lock-in and lack of portability to other systems. Moreover, some raised doubts over the novelty of customized versions, believing many to be reworked versions of prompts that are already publicly available.

Users also expressed frustration with the rollout, particularly paying users irked by gradual access and bugs. The limited visibility into OpenAI’s inner workings made users feel more in the dark. Critics argued more transparency is required regarding training data provenance and differences between old and new models.

In summary, reactions spanned from hopeful to dubious, often recognizing the immense potential while harboring reservations about OpenAI’s strategy. The community appears hungry for ongoing AI progress but wary of uncontrolled consolidation of power, signaling the importance of openness and competition alongside capabilities. Harnessing AI safely for the benefit of all will likely entail a careful balancing of competing interests.

Announcement Cheat Sheet

Here are the specific features announced this morning:

Models

Larger Context Windows:

GPT-4 Turbo supports 128K context size
GPT-3.5 Turbo increased to 16K context size
Significance: Allows understanding and generation of longer text.

GPT-4 Turbo:

Upgraded GPT-4 with 3x cheaper input pricing, 2x cheaper output pricing, improved instruction following
Significance: Major increase in capabilities and affordability.

GPT-4 Turbo with Vision

GPT-4 Turbo can now process visual inputs.
Examples: image captioning, analysis.
Significance: Multimodal understanding and generation.

Tools

Assistants API:

Enables building multi-turn AI assistants.
Examples: data analysis, coding, planning apps.
Significance: Streamlines assistant development.

Fine-Tuning:

Experimental GPT-4 fine-tuning access, expanded GPT-3.5 fine-tuning capabilities.
Examples: Industry/task-specific customization.
Significance: More tailored performance.

Custom Models:

Program for training custom GPT-4 models tailored to unique domains and massive datasets.
Examples: Specialized enterprise applications.
Significance: Customization of LLMs.