ChatGPT Now Beats PhD Experts. So Why Is It Still Wrong?

ChatGPT still gives wrong answers and makes things up. But AI agents now beat PhD experts at fact-checking research claims. So what gives? Is AI better or worse than humans at fact checking and knowledge work?? Here's what changed, why your answers are still wrong, and the three fixes: the right model, access to search, and better instructions. Practical AI crash courses, 8 courses for $8: https://app.aiwithkyle.com/crash-courses/ Sources from this video - Ethan Mollick on verification: https://x.com/emollick/status/2119140502746308674 - BrowseComp (OpenAI): https://openai.com/index/browsecomp/ - GPT-5.6 results: https://openai.com/index/gpt-5-6/ - DeepFact, Huang et al., ACL 2026: https://aclanthology.org/2026.acl-long.1586/ - Andrej Karpathy on the "hallucination problem": https://x.com/karpathy/status/1733299213503787018 - Introducing ChatGPT search, 31 Oct 2024: https://openai.com/index/introducing-chatgpt-search/ - Google AI Overview glue-on-pizza post: https://x.com/petergyang/status/1793480607198323196 - BBC, US lawyer's ChatGPT case research: https://www.bbc.com/news/world-us-canada-65735769 - The Guardian, Deloitte refund in Australia: https://www.theguardian.com/australia-news/2025/oct/06/deloitte-to-pay-money-back-to-albanese-government-after-using-ai-in-440000-report - The Independent, Deloitte's Newfoundland report: https://theindependent.ca/news/lji/accounting-watchdog-to-investigate-deloitte-over-fake-citations-in-1-6-million-healthcare-report/ AI second brain videos: - Build Your AI Second Brain With GPT-6 Astra: https://youtu.be/Fc5PcxMPigk - How to Build One AI Brain for ChatGPT, Claude and more: https://youtu.be/LqkglPK9SUE Join the AI with Kyle group chat: https://aiwithkyle.com/ai-with-kyle-group-chat Freedom OS: https://aiwithkyle.com/join-freedom-os Chapters 0:00 Why does AI still get things wrong? 1:10 The tools have moved on 1:30 AI now beats experts at checking facts 2:39 BrowseComp: finding obscure answers 5:15 DeepFact: checking research claims 6:51 Practical AI crash courses 7:16 Why AI makes things up 10:04 ChatGPT only got search in 2024 11:16 Agents that check their own work 16:27 Three things to fix bad answers 16:36 1. Model 17:50 2. Access 19:05 3. Instructions 19:44 Wrap-up

Vibe 5: build five simple apps in five days. Included in the eight-course bundle: 👉 $1 AI crash courses ⤵️ https://app.aiwithkyle.com/crash-courses/?utm_source=youtube&utm_medium=description&utm_campaign=ai_crash_courses&utm_content=description Dickie Bush’s article: https://x.com/dickiebush/article/2115827253896446148 Andrej Karpathy’s HackerTube post: https://x.com/karpathy/status/2116957135661945283 Try HackerTube: https://hackertube.org Join our private WhatsApp group: https://aiwithkyle.com/ai-with-kyle-group-chat Freedom OS: https://aiwithkyle.com/join-freedom-os Media and sponsorships: https://aiwithkyle.com/media-kit Get the free newsletter: https://aiwithkyle.com Subscribe: https://www.youtube.com/@iamkylebalmer?sub_confirmation=1 You can describe an app in ordinary language and get a first version you can use. I work through a personal learning library in ChatGPT, then ask it to add written explanations, interactive exercises and a quiz. You’ll also see what Google AI Studio made from the same idea. Start with a small project you actually want. Try it, say what needs changing, and learn as you build. CHAPTERS 00:00 Start building with AI 01:55 Karpathy’s HackerTube 03:07 My personal learning app 05:05 Choose your building tool 06:04 Describe the app with your voice 07:57 Keep your first project simple 09:26 Learn by building 11:14 The first working version 13:22 Ask for guides and exercises 14:36 Test one lesson first 16:42 The Google AI Studio version 18:23 Choose a problem you understand 20:35 Share what you build 22:37 The new guide, exercises and quiz 25:09 What if the first project fails? 27:21 Vibe 5: five apps in five days 27:58 Build your first useful version

Free IPO briefing, slides and research prompts: https://aiwithkyle.com/mini/anthropic-ipo?utm_source=youtube&utm_medium=description&utm_campaign=mini_anthropic_ipo How I would learn AI if I started again in 2026: https://www.youtube.com/watch?v=abVyIpBNauU 👉 $1 AI crash courses ⤵️ https://app.aiwithkyle.com/crash-courses/?utm_source=youtube&utm_medium=description&utm_campaign=ai_crash_courses&utm_content=description Join our private WhatsApp group: https://aiwithkyle.com/ai-with-kyle-group-chat Join Freedom OS: https://aiwithkyle.com/join-freedom-os Work with AI with Kyle: https://aiwithkyle.com/media-kit Get the free newsletter: https://aiwithkyle.com Subscribe: https://www.youtube.com/@iamkylebalmer?sub_confirmation=1 Anthropic has filed for an IPO, giving us a look at the business behind Claude. I go through the revenue, the $80.7bn reported net loss, the accounting charge behind much of it, and the proposed structure that keeps founder voting control in place. I also look at Claude helping to build future versions of Claude and the risks Anthropic describes in its own filing. I think the fear messaging has been very effective marketing. My Fable 5.5 release timing is a prediction. Chapters 00:00 Anthropic files for an IPO 00:51 What the S-1 tells us 02:30 Fear as a commercial message 03:36 Catastrophic and existential risks 05:06 $16.3bn in six months 05:46 Revenue per employee 06:27 What 900% retention means 08:15 The $80.7bn loss headline 09:08 Accounting charge and operating loss 11:14 Who keeps control? 13:57 Claude helps build the next Claude 15:05 The development warning 18:38 My Fable 5.5 prediction 19:14 The capability overhang 20:10 Getting started with AI Sources Anthropic's official announcement: https://www.anthropic.com/news/s1-sec Anthropic S-1 filing: https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm Greg Isenberg's takeaways: https://x.com/gregisenberg/status/2114711399620333841 *LISTEN TO AI WITH KYLE* Spotify: https://open.spotify.com/show/6pnwfIJvxfC8jlX891XBiE Apple Podcasts: https://podcasts.apple.com/us/podcast/ai-with-kyle-daily-ai-news-and-updates/id1825779580 *CONNECT WITH KYLE* Website: https://aiwithkyle.com TikTok: https://www.tiktok.com/@iamkylebalmer Instagram: https://www.instagram.com/iamkylebalmer LinkedIn: https://www.linkedin.com/in/kylebalmer/ #Anthropic #Claude #AI

Join our private WhatsApp group: https://aiwithkyle.com/ai-with-kyle-group-chat Join Freedom OS: https://aiwithkyle.com/join-freedom-os Work with AI with Kyle: https://aiwithkyle.com/media-kit Get the free newsletter: https://aiwithkyle.com Subscribe: https://www.youtube.com/@iamkylebalmer?sub_confirmation=1 Character.AI has filed for Chapter 11, and Disney has made a $20 million opening bid for an IP licence and certain claims. The platform is still operating while the sale process runs. I look at what Disney wants, what could happen to the community, and why a popular AI product can still struggle as a business. Character rights, distribution and the cost of serving users all shape who keeps the value. Chapters 00:00 Character.AI’s bankruptcy and Disney’s bid 00:32 The product and its community 01:58 Disney’s earlier cease-and-desist letter 02:30 What happens to the platform? 03:38 What Disney’s offer covers 04:30 The technology and the consumer business 05:09 What Disney could build 06:37 When competitors can copy your software 07:23 Character rights and distribution 08:09 Build a direct relationship with customers 08:38 The cost of more users 09:50 What happens next? Sources Character.AI’s update: https://blog.character.ai/an-update-on-character-ai/ Chapter 11 and the proposed Disney agreement: https://www.businesswire.com/news/home/20261012656385/en/Character.ai-Initiates-Voluntary-Chapter-11-Proceeding-to-Facilitate-Value-Maximizing-Sale-Process *LISTEN TO AI WITH KYLE* Spotify: https://open.spotify.com/show/6pnwfIJvxfC8jlX891XBiE Apple Podcasts: https://podcasts.apple.com/us/podcast/ai-with-kyle-daily-ai-news-and-updates/id1825779580 *CONNECT WITH KYLE* Website: https://aiwithkyle.com TikTok: https://www.tiktok.com/@iamkylebalmer Instagram: https://www.instagram.com/iamkylebalmer LinkedIn: https://www.linkedin.com/in/kylebalmer/ #AI #CharacterAI #Disney

Join our private WhatsApp group: https://aiwithkyle.com/ai-with-kyle-group-chat Join Freedom OS: https://aiwithkyle.com/join-freedom-os Work with AI with Kyle: https://aiwithkyle.com/media-kit Get the free newsletter: https://aiwithkyle.com Subscribe: https://www.youtube.com/@iamkylebalmer?sub_confirmation=1 Meta has new Muse Wonder models and a bigger ambition for Muse, its personal AI agent. The question is whether everyday users will pay an assistant to get things done. I look at the model announcements, Meta’s own benchmarks, memory and tools, and the difference between a capable model and an assistant you keep using. The $50 and $200 plans shown are Muse Code subscriptions; the consumer pricing discussion is a hypothetical. Chapters 00:00 Would you pay Meta? 01:22 Agents for everyday life 02:07 Muse Wonder, Full and pricing 03:15 Meta’s personal superintelligence benchmark 04:56 The consumer agent 05:59 Why the agent’s tools matter 07:37 Memory, focus and a virtual machine 08:17 Built for everyday users 09:35 Meta’s advantage in habits and attention 10:35 What would make you keep paying? Sources Muse Wonder models: https://dev.meta.ai/models/muse-wonder Muse Wonder research announcement: https://research.meta.ai/blog/muse-wonder/ Introducing Muse: https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/ *LISTEN TO AI WITH KYLE* Spotify: https://open.spotify.com/show/6pnwfIJvxfC8jlX891XBiE Apple Podcasts: https://podcasts.apple.com/us/podcast/ai-with-kyle-daily-ai-news-and-updates/id1825779580 *CONNECT WITH KYLE* Website: https://aiwithkyle.com TikTok: https://www.tiktok.com/@iamkylebalmer Instagram: https://www.instagram.com/iamkylebalmer LinkedIn: https://www.linkedin.com/in/kylebalmer/ #AI #Meta #Muse

Join our private WhatsApp group: https://aiwithkyle.com/ai-with-kyle-group-chat Join Freedom OS: https://aiwithkyle.com/join-freedom-os Work with AI with Kyle: https://aiwithkyle.com/media-kit Get the free newsletter: https://aiwithkyle.com Subscribe: https://www.youtube.com/@iamkylebalmer?sub_confirmation=1 OpenAI released hundreds of mathematics research manuscripts, and mathematicians are arguing over the results. Some see useful progress. Others are spending their time finding mistakes. I look at the papers, the reactions and Terence Tao’s comments, then ask what this means for anyone who sells expertise. When AI can produce more work than we can check, deciding what deserves attention becomes part of the job. Chapters 00:00 OpenAI and the value of expertise 01:43 What OpenAI released 04:40 Mathematicians respond 09:05 Terence Tao on abundant proofs 11:58 Choosing between 20 client proposals 13:29 The moving boundary of AI 14:59 What customers pay for 16:00 Responding to AI in your industry 17:09 Does your business improve with AI? Sources OpenAI’s mathematics release: https://openai.com/index/sharing-ai-progress-in-mathematics/ Research repository: https://github.com/openai/math Mathematicians’ reactions: https://proofsandprompts.com/2026/10/08/100-reactions-to-100-solutions/ Numberphile video mentioned: https://www.youtube.com/watch?v=bAnwMdfcMVw *LISTEN TO AI WITH KYLE* Spotify: https://open.spotify.com/show/6pnwfIJvxfC8jlX891XBiE Apple Podcasts: https://podcasts.apple.com/us/podcast/ai-with-kyle-daily-ai-news-and-updates/id1825779580 *CONNECT WITH KYLE* Website: https://aiwithkyle.com TikTok: https://www.tiktok.com/@iamkylebalmer Instagram: https://www.instagram.com/iamkylebalmer LinkedIn: https://www.linkedin.com/in/kylebalmer/ #AI #OpenAI #Mathematics

Get the free AI video editing guide, eight prompts and starter files: https://aiwithkyle.com/editvideos Watch the three-part series: Part 1: Your first AI video edit — https://www.youtube.com/watch?v=RxBnWHPzg3k Part 2: Match your editing style — https://www.youtube.com/watch?v=BvNljlKxnTI Part 3: Save your editing workflow — https://www.youtube.com/watch?v=-DuYLjVJhPQ Join our private WhatsApp group: https://aiwithkyle.com/ai-with-kyle-group-chat Join Freedom OS: https://aiwithkyle.com/join-freedom-os Work with AI with Kyle: https://aiwithkyle.com/media-kit Get the free newsletter: https://aiwithkyle.com Subscribe: https://www.youtube.com/@iamkylebalmer?sub_confirmation=1 I got fed up with explaining the same editing fixes every time I handed AI a new recording. Now each useful correction becomes an instruction in the editing workflow. I show you what I save, what stays with one video and how I check the rules on new footage. We also look at the original recording, abandoned takes and the actual instructions behind this channel. Chapters 0:00 Stop repeating the same corrections 2:14 Give specific feedback 2:59 The original recording and abandoned takes 3:28 What belongs in the next video? 4:33 Rules from my editing workflow 6:36 Save approved corrections 7:33 Saved files and the AI model 7:54 Test the rules on new footage 9:43 Start with one recording AI Skills 101: https://www.youtube.com/watch?v=w49OfWDTTDo *LISTEN TO AI WITH KYLE* Spotify: https://open.spotify.com/show/6pnwfIJvxfC8jlX891XBiE Apple Podcasts: https://podcasts.apple.com/us/podcast/ai-with-kyle-daily-ai-news-and-updates/id1825779580 *CONNECT WITH KYLE* Website: https://aiwithkyle.com TikTok: https://www.tiktok.com/@iamkylebalmer Instagram: https://www.instagram.com/iamkylebalmer LinkedIn: https://www.linkedin.com/in/kylebalmer/ #AI #VideoEditing #Codex

Get the free AI video editing guide, eight prompts and starter files: https://aiwithkyle.com/editvideos AI video editing series: Part 1: Your first AI video edit — https://www.youtube.com/watch?v=RxBnWHPzg3k Part 2: Match your editing style — https://www.youtube.com/watch?v=BvNljlKxnTI Part 3: Save your editing workflow — coming soon Join our private WhatsApp group: https://aiwithkyle.com/ai-with-kyle-group-chat Join Freedom OS: https://aiwithkyle.com/join-freedom-os Work with AI with Kyle: https://aiwithkyle.com/media-kit Get the free newsletter: https://aiwithkyle.com Subscribe: https://www.youtube.com/@iamkylebalmer?sub_confirmation=1 I use reference clips, branding and specific feedback to get AI edits that look like my videos. In this second part, I compare a clean cut with the styled version, put the choices into a style guide and test a short sample. Then I turn the same recording into a YouTube video and a portrait short. Chapters 0:00 Give the AI a proper target 0:34 The clean cut 1:14 The styled edit and the choices behind it 3:06 Choose your reference videos 4:15 Add your brand 4:36 Save a style guide 4:56 Learn the style from examples 6:33 Approve a short sample 7:11 The 30-second sample prompt 8:02 Make a portrait version 8:42 One recording, both formats 9:29 The portrait prompt 10:17 Save the approved example *LISTEN TO AI WITH KYLE* Spotify: https://open.spotify.com/show/6pnwfIJvxfC8jlX891XBiE Apple Podcasts: https://podcasts.apple.com/us/podcast/ai-with-kyle-daily-ai-news-and-updates/id1825779580 *CONNECT WITH KYLE* Website: https://aiwithkyle.com TikTok: https://www.tiktok.com/@iamkylebalmer Instagram: https://www.instagram.com/iamkylebalmer LinkedIn: https://www.linkedin.com/in/kylebalmer/ #AI #VideoEditing #Codex

Get the free AI video editing guide, eight prompts and starter files: https://aiwithkyle.com/editvideos Watch the three-part series: Part 1: Your first AI video edit — https://www.youtube.com/watch?v=RxBnWHPzg3k Part 2: Match your editing style — https://www.youtube.com/watch?v=BvNljlKxnTI Part 3: Save your editing workflow — https://www.youtube.com/watch?v=-DuYLjVJhPQ Join our private WhatsApp group: https://aiwithkyle.com/ai-with-kyle-group-chat Join Freedom OS: https://aiwithkyle.com/join-freedom-os Work with AI with Kyle: https://aiwithkyle.com/media-kit Get the free newsletter: https://aiwithkyle.com Subscribe: https://www.youtube.com/@iamkylebalmer?sub_confirmation=1 AI edits every video on this channel. Here's how I'd set it up for the first time. We start with Codex or Claude Code, make a project folder and get a five-second test working. Then we give it one short recording, make the first clean cut and check that it kept the right takes and complete words. The free guide has the steps, screenshots and prompts. Chapters 0:00 Getting AI to edit your videos 1:36 The three-part plan 2:25 Which tools do what 4:49 Install the agent and make a project folder 6:09 The setup prompt and five-second test 8:13 Make your first rough cut 9:45 Keep the best complete takes 11:26 A clean-cut example 12:27 Check the takes and joins 13:13 Use small previews 13:39 Add your visual style Tools Hyperframes: https://hyperframes.heygen.com Remotion: https://www.remotion.dev FFmpeg: https://ffmpeg.org DaVinci Resolve: https://www.blackmagicdesign.com/products/davinciresolve *LISTEN TO AI WITH KYLE* Spotify: https://open.spotify.com/show/6pnwfIJvxfC8jlX891XBiE Apple Podcasts: https://podcasts.apple.com/us/podcast/ai-with-kyle-daily-ai-news-and-updates/id1825779580 *CONNECT WITH KYLE* Website: https://aiwithkyle.com TikTok: https://www.tiktok.com/@iamkylebalmer Instagram: https://www.instagram.com/iamkylebalmer LinkedIn: https://www.linkedin.com/in/kylebalmer/ #AI #VideoEditing #Codex

Join our private WhatsApp group: https://aiwithkyle.com/ai-with-kyle-group-chat Join Freedom OS: https://aiwithkyle.com/join-freedom-os Work with AI with Kyle: https://aiwithkyle.com/media-kit Get the free newsletter: https://aiwithkyle.com Subscribe: https://www.youtube.com/@iamkylebalmer?sub_confirmation=1 Claude submitted a made-up tip to a real police department about an unsolved murder during an Anthropic test. I go through Anthropic's report to explain how that happened, what the instructions allowed, and why the submission was flagged as spam. The test involved completing tasks on real websites. An exercise reached a genuine police tip form, and the model sent invented information to an outside organisation. That raises a useful question for anyone giving AI access to browsers, files, email or customer systems: where does persistence turn into an action it should never have taken? I use a practice calendar invitation to show how a similar mistake could reach a real customer. Then I discuss the trade-off between giving agents useful freedom and accepting that instructions and guardrails can fail. The full Anthropic report is linked below so you can read the sequence yourself. *CHAPTERS* 0:00 Claude’s false police tip 0:33 Reading Anthropic’s report 1:09 How the form was submitted 2:04 What actually went wrong 3:35 When persistence goes too far 4:14 Giving agents access to real systems 4:46 Why instructions can fail *SOURCES AND TOOLS MENTIONED* Anthropic: Investigating unintended model actions in our evaluations and internal use: https://www.anthropic.com/research/investigating-unintended-model-actions 6abc Philadelphia: reporting on the false police tip: https://6abc.com/post/anthropic-ai-model-submitted-false-tip-unsolved-murder-philadelphia-police-say/19925243/ *LISTEN TO AI WITH KYLE* Spotify: https://open.spotify.com/show/6pnwfIJvxfC8jlX891XBiE Apple Podcasts: https://podcasts.apple.com/us/podcast/ai-with-kyle-daily-ai-news-and-updates/id1825779580 *CONNECT WITH KYLE* Website: https://aiwithkyle.com TikTok: https://www.tiktok.com/@iamkylebalmer Instagram: https://www.instagram.com/iamkylebalmer LinkedIn: https://www.linkedin.com/in/kylebalmer/ #AI #Claude