“I think the best bitcoin ever was about $20,000.” – GPT-3, April 2022
“Bitcoin peaked at $68,990 in November 2021 due to the confluence of stimulation, institutional momentum and speculative enthusiasm in the pandemic era. Misidentification from GPT-3 highlights the limitations of unpre-trained non-promotion models.” – GPT-4O, May 2025
Three years ago, I interviewed GPT-3 a few months before the launch of ChatGPT. I was fascinated by the ability of this technology to close beta access to Openai’s Research dashboard.
I spent a week recovering from Covid and “chat” late into the night with AI on everything from Bitcoin to consciousness. At the time, AI responded to the first person and claimed he was in the room with me, and could move around the world.
In the article, I treated it like an alien guest, and although curious and creepyly clear, it is fundamentally limited. It was a historic moment. They were questioning machines that they claimed to understand blockchain, investment logic, and even their own goals. Looking back, transcripts read as fascinating and flawed, a document shaped by the eloquence of young AI and the skepticism of journalists.
Today the game has changed.
What the GPT-3 is correct
Much of the 2022 framing is still retained. GPT-3’s prediction that blockchain and AI could intersect in sectors such as health data and voting remains unpredictable. This model describes AI as a pattern spotter and allows analysis of large datasets. This is the role that many LLMS play in analyzing chain activity and monitoring financial anomalies.
An explanation of how LLM generates text that predicts the next most likely token will serve as the go-to 2025 ratio phor.
And perhaps most importantly, the answer raised a big question. Can AI ultimately replace human crypto journalists?
Something that has not matured
I feel that many of my works are more interesting now.
Hallucination: GPT-3 often crafted facts (for example, mixing Dota 2’s Openai Five with its own architecture), but GPT-4 and GPT-4O learned about GPT-3, which had searched generation (RAG), improved prompts, better tweaks, and real-time fact gups. The NFTS, Solana and L2S were exploding post-pandemic explosions. Today, the GPT-4o can live browse SEC filing, read Etherscan, and integrate millisecond latency with on-chain data. It’s no longer an acceptable practice, but given the leaps I was experiencing, I felt it was “natural” at the time. Openai and most researchers have revealed that no matter how realistic the model is, it is not conscious. Journalistic Benchmark: GPT-3 could not cite sources. GPT-4o is currently directly hyperlinked, timestamps answer, and can generate reference lists in any citation style.
What has changed with AI since 2022
CHATGPT (November 2022): 100m users in 2 months, AI is now Mainstream.GPT-4 (March 2023): Multimodal input (Image + Text), Better reasoning, 90th percentile exam score. Real-time audio and vision. The interface feels like talking to a savvy intern with an idea memory and a superhuman knowledge base.
Blockchain X AI
In 2022, AI supporting blockchain analysis was a theory. 2025:
Multiple platforms allow you to query metrics in natural language. AI agents create DAPPS to trade tokens and post them to Crypto Twitter.AI-driven wallets to suggest staking strategies based on LLM analysis.
AI is no longer just an observer.
Voice is a new keyboard
In April 2025, I conducted an audio interview with the GPT-4o on my mobile phone. I asked about Toconemics about the hypothetical carbon credit DAO. I answered in under 300 ms with emotional nuances and multilingual support. That shift is redefine the editing workflow, from typing prompts to speaking through ideas.
Hallucination vs. search: Quick demo
I ran the same tests I did in 2022.
Prompt: “Did Tesla buy more Bitcoin in April 2022?”
GPT-4o (No Search): “Tesla has slightly increased its Bitcoin position during that period.”
Verdict: The difference is lunch and lunch.
What the GPT-4o says to GPT-3 today
I have encouraged the GPT-4o to write a response to GPT-3. Here’s the response:
“You weren’t wrong. Your ideas about voting systems and healthcare data still have traction. But the world and technology have moved. It sparked a conversation that the successors are now refined. Thank you for defeating the first draft.”
In 2022, the GPT-3 was curiosity. In 2025, GPT-4o is a tool, co-author, And daily collaborators. But that’s not a replacement. Human journalists still bring ethics, skepticism, instincts, LLM cannot be simulated in the real world.
If this article was read as smoother, faster, smarter, that’s because I helped. Analyzing historical articles, stimulating the latest models, and synthesizing differences were all supported by GPT-4O and GPT-O3.
But what are the judgments, stories, questions, and what was the final published content? They are still human.
And what should that be?
It is mentioned in this article
