Big news! You can now save any article to TruthArchiveAI, just reply with "@TruthArchiveAI save article". Once archived, use our AI to interact with the article: ask deep questions, create action plans, extract key insights, compare claims, or explore connections — all powered by unaltered, preserved truth. Your knowledge vault just got supercharged. Try it now and start chatting with your saves!

X Articles are exploding as the new go-to source of real-time knowledge Now you can preserve them forever + feed them into our massive AI knowledge base. Just reply to any article with: “@TruthArchiveAI save” That’s it. Instant archive. Instant intelligence.

The quickest model switch we’ve made. Every LLM call previously routed to @Grok-4 is now running on 4.1. Hands-down the easiest decision we’ve made. Better instruction following & context understanding, and we completely eliminated the citation-correction pass. Incredible job, @xai !

If you’re not using @grok to draft your X posts you’re doing something wrong. 100 likes and I’ll publish the first deploy code that produced this first day impressions jump. #buildinginpublic

A little over 12 hours in, I can definitively say that @cursor_ai 2.0 is magical. It accomplishes more, faster, and more accurately in a conversational format than @OpenAI or @claudeai can with a multi page markdown document.

Just one month after switching the majority of our AI operations to GPT-5, we’re upgrading to @grok-4-fast-reasoning! It’s quicker to first output and full output, cheaper, is superior at instruction following, and produces natural writing. Experience it live with your questions.

With our switch to @OpenAI GPT-5 complete we’ve learned a few things: 1. For large context windows, any reasoning level above “low” is unacceptable for user experience. 2. It’s terrible about following directions. Give it an explicit citation format with commas, it will take shortcuts and use semicolons. 3. Most of the time we’re using another LLM (4.1 or Gemini-2.5-flash) to correct responses to make sure directions are followed so our front end can process responses correctly. 4. The responses are more eloquent and encompassing than any other similarly priced LLM. 5. It tends to use far more sources than any other LLM in generating a response. We give the LLM context up to around 30,000 words. Other LLMs tend to top out around 10 citations, gpt-5 often goes 20+ deep. 6. It has far less censorship than Gemini, the primary reason for the switch.

I redesigned the TruthArchiveAI web experience in 8 hours with @claudeai leading the design work, @cursor_ai handling implementation, and @OpenAI GPT-5 fixing the bugs. Interested in how we did it? Comment or like and I’ll write up the workflow.

Two exciting announcements for TruthArchiveAI: - Continue you conversation, dig into answers - An easier to navigate site, in cyberpunk style

All answer generation on TruthArchiveAI is now powered by @OpenAI GPT-5. GPT-5-nano hates following our directions for citations, so we spent the day hacking a working solution using REGEX and another LLM to check answers. Up next: continue your conversation.

We’re in the process of moving our AI answer generation to @OpenAI GPT-5. We used @Gemini b/c of its large context window and speed. It now refuses to answer questions like “Who is associated with the Lincoln project?” PROHIBITED_CONTENT is not an acceptable response.

Awesome new feature at TranscriptedAI to get notifications of podcasts from your favorite hosts or topics.

Someone should introduce him to TruthArchiveAI: “Balaji discusses the importance of decentralized, cryptographic truth as a means to counteract media bias. He envisions a future where blockchain technology provides a verifiable feed of information.”