Hashtag Trending

Project Synapse Weekend: When Cheap Local Models Hallucinate + Anthropic's Fable-5 Controversy
Jim Love and Marcel host the Project Synapse weekend edition without Jon Pinard, discussing Jim's three-month experiment building his Nexus AI agent. Trying to avoid expensive API calls, Jim switched to a small open-source Gemma model and found that limited context windows and growing memory files led to severe hallucinations and corrupted "memories," suggesting he may need periodic context compression or paid access to larger models. They explain quantization and compare AI memory to human memory reconstruction, then pivot to Anthropic's newly released Fable-5 (a guardrailed Mythos-5 variant) via Project Glasswing: powerful but slow, expensive, temporary for subscribers, and controversial for silently downgrading or misdirecting users. They also discuss AI-written apps replacing utilities, debates over AI creativity, Canadian proposals on youth social media and humanlike AI, information control by big tech, job disruption risks, and a case where lawyers cited AI-fabricated precedents in court.
00:00 Weekend Edition Kickoff
00:27 Cheap Model Reality Check
03:37 Quantization and Brain Metaphors
08:16 Nexus Agent Memory Spiral
13:26 Compressing Context Strategy
17:17 Conversational Programming Demo
23:52 What Is Fable 5
26:52 Fable Controversy and Guardrails
30:30 AI Feels Human and New Rules
34:21 LLMs Mirror You
35:29 Kids and Age Gates
36:24 Dumbed Down Society
39:07 Extraneural Fan Site Backlash
43:33 AI Art and Creativity Lines
48:17 Waterloo and Mennonite Tech
53:05 Weaponized Narratives
58:30 Billionaires and Ubiquity
01:01:19 Jobs and AI Economics
01:04:15 Lawyers Citing Fake Cases

Jim Love hosts Hashtag Trending for Friday, June 12, 2026, with a special look at a week of technology stories that range from bizarre to genuinely concerning.
It starts with a personal experience involving a Bose warranty that appeared to vanish after being successfully registered, raising questions about what happens when automated systems become the final authority and no one seems empowered to challenge them. That leads into a discussion of a recent German court ruling holding Google responsible for inaccurate AI-generated summaries and the growing push in Europe toward liability for software mistakes.
Other stories include reports that Meta is housing AI infrastructure in giant tent-like structures to get computing capacity online faster, a North Carolina software engineer who reportedly received a religious exemption from using AI at work, the NSA's reported use of Anthropic's Mythos cybersecurity AI despite tensions elsewhere in Washington, the UK's decision to help workers worried about AI-driven job losses with yet another AI tool, and a courtroom drama where lawyers on both sides submitted AI-generated legal citations so flawed that a judge cancelled the trial.
Chapters
00:00 Weekly Weirdness Intro
00:32 Bose Warranty Vanishes
06:00 Google AI Liability Ruling
08:20 Meta AI Tent Cities
10:32 Religious Exemption From AI
12:43 Anthropic And The NSA
15:02 UK Jobseekers Get More AI
16:08 Lawyers Caught By Hallucinations
18:21 Wrap Up And Weekend Plug

Europe is accelerating its push for digital sovereignty with the launch of EuroOffice, a cloud-based alternative to Microsoft 365 backed by German hosting giant IONOS and other European technology companies. But the launch has sparked controversy, with LibreOffice accusing the project of reinforcing Microsoft's document standards while critics question its roots in a fork of OnlyOffice.
Meanwhile, new data suggests Google's AI Overviews are dramatically accelerating the rise of "zero-click" searches. Nearly 69 percent of Google searches now end without users visiting another website, raising concerns for publishers, online merchants and the growing industry of search engine optimization firms now pivoting toward Answer Engine Optimization, or AEO.
Meta faces renewed criticism after former Facebook executive Sarah Wynn-Williams was effectively silenced from promoting her memoir Careless People. The dispute raises questions about whistleblower protections, corporate power and the role of a company that controls a significant share of how people communicate and consume news.
And finally, a German court may have delivered one of the most important AI rulings to date. Rejecting Google's defence that users understand AI can make mistakes, the judges ruled that people trust AI-generated answers precisely because they expect them to be useful. The decision could have major implications for whether AI companies can be held legally responsible when their systems generate false information.
In This Episode
00:00 Europe launches EuroOffice as a Microsoft alternative
02:10 Google AI Overviews drive zero-click searches to record highs
04:15 Meta's campaign against former executive Sarah Wynn-Williams
06:20 German court delivers potentially landmark AI liability ruling
Hashtag Trending is hosted by Jim Love and covers the latest developments in AI, cybersecurity, technology policy, enterprise IT and digital business.
Subscribe for daily technology news and analysis.

Jim Love covers three headlines for June 10, 2026: OpenAI has filed confidential paperwork for a US IPO, with Reuters suggesting it could come as early as September amid a broader wave of potential blockbuster listings also involving Anthropic and SpaceX; the report reviews OpenAI's shift from non-profit to commercial powerhouse, prior governance turmoil around Sam Altman, planned conversion to a public benefit corporation, and disputed private-company revenue/user figures compared with Anthropic. Meta is aggressively rebutting a Wired report that found facial-recognition references in smart-glasses software that reportedly disappeared after inquiry, amid rising scrutiny of always-on recording and proposed notice laws. Kevin O'Leary apologized for missteps around Utah's Wonder Valley AI data center plan, which has been cut roughly in half after local opposition focused on water transfers, power, costs, and environmental impacts.
00:00 Headlines and intro
00:29 OpenAI files for IPO
02:14 OpenAI vs Anthropic numbers
03:35 Meta denies facial recognition
04:50 Smart glasses privacy backlash
05:58 Kevin O'Leary data center apology
07:02 Wonder Valley water and power fight
07:49 Lessons for AI infrastructure
08:54 Wrap up and support the show

Jim Love covers four headlines: hackers exploited Instagram's AI support bot to hijack over 20,000 accounts by abusing account recovery and password reset links, prompting Meta to disable the tool, remove faulty code, and add enhanced protections. A UN University report warns AI's environmental footprint extends beyond carbon, projecting data centers could consume 945 TWh annually by 2030 and highlighting growing demands for electricity, cooling water, land, and minerals, amid political backlash to data center incentives. A UK government review found false information from a Microsoft Copilot hallucination and other inaccuracies were included in West Midlands Police materials, pointing to failures in review and validation. CBC News also identified at least 14 foreign-linked Facebook accounts posing as Albertans in separatist groups, raising concerns about deceptive political participation and platform responsibility.
00:00 Today's Tech Headlines
00:36 Instagram Bot Account Hijacks
02:03 AI Agents Security Lessons
03:12 UN Report AI Resource Footprint
05:38 Copilot Hallucination Police Report
08:11 Fake Albertans in Facebook Groups
11:04 Wrap Up and Support the Show