Hands On Testing! Open AI's New "GPTs" & ChatGPT Update!

ChatGPT’s latest update streamlines model switching and folds more multimodal tools into a single workflow—while OpenAI’s new GPTs feature promises a wave of purpose-built assistants. The most immediate change is interface-level: users no longer juggle a GPT 3.5 vs GPT 4 selector bar. Instead, a simple dropdown toggle lets ChatGPT automatically switch models for different tasks, including the newly highlighted GPT-4 Turbo (described as the “turbo 128k” variant). Early hands-on testing suggests GPT-4 Turbo is faster than the older GPT-4 experience, though heavy demand can still slow responses inside ChatGPT.

Beyond speed, the update makes prompt iteration and sourcing easier. Hovering over prior messages reveals a drawing/editing icon that allows users to retype and resubmit prompts without starting fresh chats—useful when refining wording. When responses include citations, hovering surfaces the source site’s logo and provides direct links, keeping web research and answer context in the same place. The workflow also supports “fact grounding” behavior: when asked for goldfish facts, the assistant begins by searching the web, and users can adjust instructions (e.g., requesting memory-only rather than web search).

Multimodal capabilities are treated as first-class features rather than separate experiments. Image generation with DALL·E is integrated directly into ChatGPT, including a workflow where users upload a photo and ask for a DALL·E recreation. During testing, DALL·E generation sometimes fails or stalls under load, but regeneration typically succeeds—producing an example lemon character in a pop-art style. File uploads also work inside the chat: a user uploads a document (a Wikipedia page about Squidward Tentacles), and a custom GPT reads and summarizes it, then applies that content to tasks like negotiation strategies.

OpenAI’s bigger platform shift is “GPTs,” a beta feature for creating custom versions of ChatGPT tailored to specific purposes. The account area includes “My GPTs,” plus a “Create a GPT” option (beta, with availability described as coming in the coming weeks). OpenAI also provides a set of pre-made GPTs—some framed as legacy-style (e.g., DALL·E, data analysis, and ChatGPT classic) and others as specialized assistants such as Game Time, The Negotiator, Creative Writing, Tech Support Advisor, Coloring Book Hero, and a meme-focused helper. The transcript emphasizes that the real value may arrive when the community can publish and share GPTs through a future store.

Hands-on tests show how these GPTs behave differently: Game Time explains Monopoly for a child; The Negotiator generates a character-specific negotiation plan using uploaded Squidward material; Coloring Book Hero turns ideas into stylized images and can transform an uploaded image into a coloring-book-like version (with at least one request denied). The Tech Support Advisor offers practical troubleshooting steps for an XLR microphone noise issue, including checks for cable connections, gain levels, phantom power, electrical interference, and potential ground-loop causes. Overall, the update pairs a simpler interface with faster GPT-4 Turbo performance and a growing ecosystem of specialized assistants—positioning ChatGPT as a more modular toolset rather than a single generic chatbot.