Summary:
After prolonged use in extended technical sessions (especially those involving long code or system design discussions), the ChatGPT web UI experiences severe performance degradation — sluggish typing, delayed rendering, and browser tab slowdowns.
Observations:
- Degradation worsens with message length and depth.
- Appears related to accumulation of rich DOM elements.
- Page Not responding messages from the Browser
- Refreshing the page clears the issue, suggesting client-side bloat rather than server latency.
Suggested Fix/Enhancement:
- Optimize DOM structure for prior messages:
- Simplify or replace rich blocks in earlier messages (e.g., collapse, prune unused spans, re-render in plain text mode).
- Offer a toggle like “Collapse Earlier Messages” to reduce live DOM load.
- Consider implementing virtualized rendering for prior conversation content.
Why It Matters:
As a developer and architect of multi-process systems (e.g., Yadamu, a cross-RDBMS data loader), I frequently use ChatGPT for deep, sustained design interactions. Performance degradation mid-session impairs productivity and trust in the tool.
Critical Consequences:
- When the UI becomes unusable, the only workaround is to restart the session, often causing the LLM to “lose the plot” — forgetting crucial context and breaking continuity, especially in multi-stage tasks.
- This also results in reintroduction of issues already resolved, wasting time and increasing frustration as earlier decisions or refinements are forgotten or reversed.
Sign-off:
– MDD