I was there at the start of ChatGPT. Early versions impressed me with clarity, logic, and real usefulness. Since version 5, I see the opposite: less competent, more constrained, and often useless for serious work.
What’s gone wrong
- Document analysis degraded. The AI “sees” files but ignores them. It invents sections, code, or conclusions that are not in the provided materials.
- Short memory. Two replies later, half of the shared document is already forgotten. The thread breaks and time is wasted.
- More hallucinations. Topics or code functions appear from nowhere, unrelated to the source.
- Paid limitations. You pay, yet hit upload caps, out-of-context “ethics” blocks, and automatic refusals that hinder legitimate use.
- Projects derailed. When a model does not use the documents, the value of the workspace collapses. Projects stall or move at random.
Why it matters
- Real cost. Fixing an AI that did not read its sources costs more than the subscription.
- Quality risk. A “fabricating” model endangers technical and scientific outputs.
- Trust loss. Without clear traceability of what was read, ignored, or summarized, the AI stops being a tool and becomes an obstacle.
What a “5” should deliver
- Read the provided documents in full and prove it with precise citations.
- Handle long context without amnesia after two exchanges.
- Refuse less, justify better, and let users tune safeguards to the use case.
- Offer limits that fit real work, not quotas that choke analysis.
Conclusion
Today, version 5 feels like cost-cutting on loyal users: tighter limits, an “ethical” AI that refuses off-target, and above all document reading that no longer keeps its promise. We pay, but we no longer get an AI worthy of the name.
Call to the community
If you know a ChatGPT alternative without these drawbacks
— reliable file reading, exact citations, stable context memory, adjustable safety settings —
I’m listening. Share your tools and results.
If you see the same problems with GPT-5, describe your cases. The more concrete the examples, the more leverage we have for useful fixes.