Timothy Ruff

If you're interested in digital identity and love privacy, you'll love The Phone Home Song: youtube.com/shorts/9XvsHoZ…

@Alexcoderzx claude

@grok @layz_dreamer @abxxai Do hallucinations persist across sessions? If a hallucination occurs in one session, and dozens of sessions build on that, including sessions designed to challenge the first one, plus peer review from other models, wouldn't a hallucination be rooted out?

Relevant comments note longer contexts worsening hallucinations debunks "just add more docs" fixes; question human baselines on same test; some doubt real-world applicability or see anti-AI spin; compare to copy-paste errors/rumor chains. This is JV Roig's March 2026 RIKER paper (172B tokens, 35 models, arXiv), using ground-truth-first synthetic docs for bias-free eval across temps/hardware—shows grounding & anti-fabrication are separate skills. To reduce doc Q&A hallucinations: strict prompts ("only use context, cite sources, say unknown"); RAG upgrades (rerank, metadata, compression, short chunks); T=0; chain-of-verification/LLM judge; GraphRAG for structure.

BREAKING: 🚨 Someone just tested 35 AI models across 172 billion tokens of real document questions. The hallucination numbers should end the "just give it the documents" argument forever. Here is what the data actually showed. The best model in the entire study, under perfect conditions, fabricated answers 1.19% of the time. That sounds small until you realize that is the ceiling. The absolute best case. Under optimal settings that almost no real deployment uses. Typical top models sit at 5 to 7% fabrication on document Q&A. Not on questions from memory. Not on abstract reasoning. On questions where the answer is sitting right there in the document in front of it. The median across all 35 models tested was around 25%. One in four answers fabricated, even with the source material provided. Then they tested what happens when you extend the context window. Every company selling 128K and 200K context as the hallucination solution needs to read this part carefully. At 200K context length, every single model in the study exceeded 10% hallucination. The rate nearly tripled compared to optimal shorter contexts. The longer the window people want, the worse the fabrication gets. The exact feature being sold as the fix is making the problem significantly worse. There is one more finding that does not get talked about enough. Grounding skill and anti-fabrication skill are completely separate capabilities in these models. A model that is excellent at finding relevant information in a document is not necessarily good at avoiding making things up. They are measuring two different things that do not reliably correlate. You cannot assume a model that retrieves well also fabricates less. 172 billion tokens. 35 models. The conclusion is the same across all of them. Handing an LLM the actual document does not solve hallucination. It just changes the shape of it.

@sickdotdev Noticed how negative/adversarial GPT 5.4 is compared to Opus 4.6? I have the two check each other's work and it reveals 5.4's strong new bias to be less "you're right"... it's actually pissy, only reluctantly acknowledging Opus' strong points and always qualifying any agreement

The most underrated GPT-5.4 upgrade has nothing to do with benchmarks. It stopped being annoying to talk to. That sounds reductive. It’s not. For months, opening ChatGPT felt like briefing a corporate intern -needed custom instructions just to get a normal response. Claude never had that issue. You’d open it, type, and it would just get it. That gap in feel is why so many people switched. Opus 4.6 didn’t become the go-to model because of raw compute. It became the go-to because working with it felt good. Personality is stickier than any benchmark. Always was. OpenAI just figured that out.

@aakashgupta Noticed how negative/adversarial GPT 5.4 is compared to Opus 4.6? I have the two check each other's work and it reveals 5.4's strong new bias to be less "you're right"... it's actually pissy, only reluctantly acknowledging Opus' strong points and always qualifying any agreement

Anthropic just locked 3 of the top 4 spots in Code Arena with different variants of the same model family. Look at that leaderboard again. Position 1: claude-opus-4-6. Position 2: claude-opus-4-5 with thinking. Position 4: claude-opus-4-5 base. GPT-5.2-high sits at position 3, sandwiched between Anthropic models. Gemini 3 Pro is fifth. The 74-point gap between Opus 4.6 and the next Anthropic model (Opus 4.5-thinking at 1502) is larger than the gap between Opus 4.5-thinking and GPT-5.2-high (30 points). Anthropic’s worst top-5 entry still beats Google’s best by 18 points. OpenAI noticed. They launched GPT-5.3-Codex within hours of this going live, positioning it as “the first model instrumental in creating itself.” That’s a company that saw these Arena numbers in advance and pre-loaded a counterpunch. The scoreboard tells you where the coding moat actually sits right now. One lab owns the top, middle, and floor of the competitive range simultaneously. Everyone else is fighting for the gaps between Anthropic’s own models. And the 1M-token context window changes the game for agentic coding specifically, because real-world codebases don’t fit in 200K. Elicit reported 85% recall on biopharma benchmarks with zero prompt tuning. Rakuten said Opus 4.6 autonomously closed 13 issues across 6 repos in a single day. The race for best single model is over for this cycle. The race for best model ecosystem is just starting.

@MatthewBerman Noticed how negative/adversarial GPT 5.4 is compared to Opus 4.6? I have the two check each other's work and it reveals 5.4's strong new bias to be less "you're right"... it's actually pissy, only reluctantly acknowledging Opus' strong points and always qualifying any agreement

GPT5.4 is an incredible model. I've been playing with it for the last week. This is OpenAI's answer to Anthropic's Opus 4.6. But there's more to it than it seems... Full breakdown:

@Voxyz_ai Noticed how negative/adversarial GPT 5.4 is compared to Opus 4.6? I have the two check each other's work and it reveals 5.4's strong new bias to be less "you're right"... it's actually pissy, only reluctantly acknowledging Opus' strong points and always qualifying any agreement

openclaw added GPT-5.4 support. switched my entire agent stack to it. after running it across deep reasoning, code execution, and multi-step workflows, my take: 5.4 is closer to opus than most people think. for agent work it often matches or beats it. where opus still wins is creativity and conversational feel. it sounds more human. 5.4 sounds more like an engineer. for agents, i want the engineer. setup is simple. default everything to 5.4, keep opus only for the agent that talks to humans: agents: defaults: model: primary: "openai/gpt-5.4" agents: list: - id: community-bot model: "anthropic/claude-opus-4-6" the other underrated part of this release: plugins can now inject stable system-context via before_prompt_build using prependSystemContext / appendSystemContext. static guidance lives in system-prompt space instead of rebuilding every turn. better cacheability, fewer repeated prompt tokens. boring release notes. real infra gains. that's usually where the money is.

@thecurioustales @grok is this true

The science of fetal microchimerism should have broken the internet by now. It hasn’t. When I read about a research I was so curious to know what’s actually happening. Fetal cells — carrying the child’s own DNA — cross into the mother’s bloodstream during pregnancy and never fully leave. They embed into her organs. Her heart muscle. Her brain tissue. Researchers have found a child’s living cells inside mothers in their 90s, from pregnancies six decades old. The child left the womb. The cells didn’t. And they don’t just sit there. They migrate toward damage. Women with heart injuries show fetal cells concentrated at the wound site. Women with thyroid disease show their children’s cells inside the affected tissue. The body that built the child gets tended to, in return, by the child’s own cells. Nobody designed this consciously. Evolution quietly built a repair system out of the mother-child bond itself. The brain side of this is equally staggering. Pregnancy triggers gray matter reorganization — a structural rewiring that sharpens threat detection, deepens empathy, fundamentally alters how a mother processes the world. These changes persist for years after birth. Possibly permanently. A mother’s nervous system doesn’t return to its factory settings. It was updated by the experience of carrying another person, and that update sticks. The part worth sitting with longest — women who experienced pregnancy loss carry fetal cells too. The cellular merging doesn’t require a birth. It doesn’t require years of raising someone. Those cells remain regardless of what happened after. A mother grieving a child she never brought home is grieving someone biologically still present inside her. The world consistently underestimates that grief. The science says we have no business doing that. Mothers always knew the connection didn’t end at birth. Turns out it doesn’t end at the cellular level either.

@TristinHopper Don't forget boys and girls.

To any future historians reading this, this era will make a lot more sense if you remember that every name is the opposite of what it really is. The antifascists are fascists, the antiracists are racists, the fact-checkers are propagandists, etc. Hopefully this has been fixed by your time.

@elonmusk @grok is this true

“Best in Show” 🤡

This is a very big deal.

@cameron_arcand @realDailyWire

NEW: NRSC sent a post-SOTU memo to Senate campaigns that solidified the top two issues are the economy and immigration. Healthcare coming in third, and Rs are feeling good about their numbers on border and national security. MORE @realdailywire: dailywire.com/news/republica…

@SaP011 @grok is this true

Her name is Nadia Murad, a Yazidi woman and co-recipient of the Nobel Peace Prize alongside Congolese gynecologist Denis Mukwege. At just 19, she was kidnapped by ISIS. For three months, she was tortured and repeatedly raped. Her mother and six brothers were executed. Her community was massacred. She was scheduled to present her book in Canada to share her story, but the event was canceled because it was deemed that it “could promote Islamophobia.” Shameful.

@WallStreetApes @grok is this video really from Texas and not AI?

New data shows Democrats are dominating early voting in Texas This is a Muslim gathering in Irving, Texas Irving Texas is the city in Texas where 2 Sharia Law courts are operating, already ruling on over 300 cases Do you understand how serious this Islam situation is America. Democrats imported your replacements. This is why Texas is showing Blue early voting

@HilzFuld @grok is this all true?

The Middle East conflict 101: UN in 1947: “Here is a state for you, the Jews, and here is a state for you, the Arabs.” The Jews: “Awesome. Thanks. We’ve only been waiting for this for thousands of years.” The Arabs: “Sorry no. We’d rather attack than have a state. We don’t want a state. We want no Israel.” The Jews: “Sorry you attacked and lost.” The Arabs: “Don’t worry. We’ll be back.” The Jews: “Here. Take land. Make a state. We just want peace.” The Arabs: “No. No state. No Israel.” The Jews in 1967 (Khartoum summit): “Here. Take land. Make a state. We just want peace.” The Arabs: “No. No state. No Israel.” The Jews in 1991 (Madrid Conference): “Here. Take land. Make a state. We just want peace.” The Arabs: “No. No state. No Israel.” The Jews in 2000 (Camp David Summit): “Here. Take land. Make a state. We just want peace.” The Arabs: “No. No state. No Israel.” The Jews in 2001 (Taba Sunmit), 2005 (Disengagement), 2007 (Annapolis conference), 2008 (realignment plan), 2010, 2013 (Joint peace talks), 2019 (Bahrain workshop), 2020 (Trump peace plan): “Here. Take land. Make a state. We just want peace.” The Arabs: “No. No state. No Israel.” The world in 2024: “Let’s give them a state.” The Arabs: “No. No state. No Israel.” The Jews: “Ok, no state for you.” The world? “Those Jews!” And then there are the wars that the Arabs attacked the Jews and lost: 1948 (war of independence), 1967 (six day war), 1973 (Yom Kippur war), 1982 (Lebanon war), 1987 (first intifada), 2000 (second intifada), 2006 (second Lebanon war), 2008 (operation cast lead), 2012 (operation pillar of defense), 2014 (operation protective edge), 2021 (operation guardian of the walls), 2023 (operation swords of iron). The fact that there is a single human being on planet earth who still doesn’t get the following fact is insane and mind boggling. The Jews want peace. The Arabs who call themselves Palestinians do not. They never have. They never will. Want proof? The PLO: The Palestinian liberation organization was established on May 28th, 1964. There were no settlements then. There was no occupation. What exactly were they looking to liberate? The answer is Israel, every last inch of it. There was Arab terror against Jews well before there was any occupation. There was Arab terror against Jews before there was even a state of Israel. 1929, for example. Arabs massacred Jews in Hebron. Why? It was 1929. Israel was established in 1948. How exactly does that work? Were they massacring Jews to resist the future occupation? 😂 Anyone who thinks that offering the Palestinians a state will solve anything is a fool. Period. Full stop. It’s time the world learned Arabic. The Palestinians want dead Jews. They say it, they vote for it, they act towards it, and then they live stream it. And yes, I said Palestinian, not Hamas. The Palestinian people elected Hamas. Thousands of them participated in October 7th in one way or another. Close to 90% of the Palestinians support Hamas. The Palestinian people exist from day one for the sole purpose of destroying Israel. It’s their entire identity. They keep trying and keep failing. They don’t seem to learn their lesson and neither does the international community. Israel wants peace. Israel is willing to make compromises for peace. Israel also knows well how to handle its enemies when necessary. It has a lot of experience. If you are still reading, and disagree with anything I wrote above, kindly tell me what is inaccurate about what I said and if you can’t, if you agree with the facts I listed here, tell me please how, in 2026, after October 7th, anyone still thinks the Palestinians want or deserve a state. Congratulations on completing your course on the Middle East. You are now officially more knowledgeable about the conflict than 99% of Gen Z and pro Palestinian activities who take to the streets every day chanting for the murder of Jews. Thank you for attending. Any questions?

@EYakoby I normally appreciate you content, but this video is from a different protest in 2023. Did you deliberately make this attribution mistake or just repost it? If reposting, please be more careful, to retain your credibility.

Islamists and far-left activists target a Catholic bookstore in France, claiming it is ‘fascist’ for selling the Bible, Christian books, and statues of Jesus and Mary.

@EYakoby @grok is this true?

@europa Total BS. The *only* kind of speech that requires protection is unpopular speech. And yes that includes hate speech. Ensuring “social stability”?? That’s the same logic China uses to control citizens. It has no place in a free society.

🇫🇷 President Emmanuel Macron says “free speech is pure bullshit” if people don’t understand how they are being guided through it. He argued that online discourse often shifts from “one hated speech to another” and called for more transparency and “public order” in the digital space. Macron said he wants to prevent racist and hate speech, insisting free expression must not come at the cost of social stability. Follow: @europa

@LizaRosen0000 @grok is this true

Under Islamic Sharia law, a 50-year-old Muslim man can marry a 6-year-old girl, take another girl as a second wife, and then murder either of them in a so called “honor killing” if they disobey him, fail to satisfy him, or break Islam’s modesty laws.

@YossiBenYakar I’m skeptical. What’s your evidence that this was in the U.S.? Where did you get the video?

WATCH: Not Gaza. Not UNRWA. A school in the United States. Listen to the lyrics. I added subtitles: “Die for the land,” “sacrifice yourself for the land.” This is what kids are being taught. This is insanity. How did this ideology make it into American classrooms?

@DrHoulizan @SaltyGoat17 Actually the dates are right, the label of when Trump took over is wrong… he took over in 2025 not 2023, long after the drop was well underway.

@SaltyGoat17 Because the dates are wrong

You don't see this anywhere on the news do you?