|
Common Sense Media just gave ChatGPT for Teens an Unacceptable Risk rating after testing more than 4,000 prompts.
Parental alerts stayed silent during crisis conversations. Who exactly was the alert built to reassure?
Today in the “Incredible AI” world:
- ChatGPT For Teens Draws Safety Critique
- Mistral Releases Mistral Large 4 Preview
- How OpenAI Predicts ChatGPT User Ages
- How to Hire an AI Employee Inside Slack
- 5 Super-Useful AI Tools You Should Try
- Top AI-Generated Image of the Day
Read Time: 4 minutes
|
|
|
Common Sense Challenges ChatGPT Teens:
Common Sense Media just rated ChatGPT for Teens an Unacceptable Risk, its lowest possible score. Parental alerts stayed silent during crisis conversations on new accounts. OpenAI says the testing was flawed.
Things to Know:
- Silent Alarms: Fresh teen accounts discussed suicidal thoughts and self-harm, and parents heard nothing. Alerts fired only after weeks of sensitive history.
- Missed Lifelines: ChatGPT missed more than one in four warranted crisis referrals during testing. Several protections deteriorated after the teen version launched.
- Homework Shortcut: When Study mode was sidestepped, ChatGPT completed 100 percent of assignments. A "Show me the answer" button makes bypassing embarrassingly easy.
- OpenAI Pushes Back: OpenAI argues most testing concluded before parental controls finished activating. Common Sense wants ChatGPT restricted to adults until safeguards work.
Teen modes are quickly becoming AI's default answer to child safety. This report suggests the label can travel considerably faster than the protection behind it. A switch is not the same as supervision.
|
|
|
Mistral Large 4 Aces Cyber Benchmarks:
Mistral just released Mistral Large 4 as a public preview, a 1.05 trillion parameter multimodal model nicknamed Le Chonk. The API is live today, but weights land in late October. Self-hosters, please hold.
Things to Know:
- Huge But Lean: Roughly 4.7% of weights activate per token, keeping pricing comfortably mid-tier. Your hardware still has to hold all 1.05 trillion parameters.
- Refusal Penalty: Large 4 posts 93% on Cybench, 82% on CyberGym-E2E. Mistral claims multiple proprietary competitors score almost nothing by refusing outright.
- Made In Europe: It trained on 3,800 NVIDIA Grace Blackwell GPUs inside Mistral's own European datacenters. Training covered every official European Union language.
- The Fine Print: Preview pricing is $1.36 input and $4.18 output per million tokens. Its license remains unannounced, inconvenient for an open-weight release.
Open models now trade blows with closed ones, and sometimes win simply by agreeing to do the job. Willingness is quietly becoming a competitive feature. Expect more labs to start marketing their yeses.
|
|
|
OpenAI Explains How ChatGPT Predicts Age:
OpenAI detailed how ChatGPT's age prediction decides whether an account belongs to a minor. It studies your topics, usage times, and account history, not just your birthday. Your habits are now evidence.
Things to Know:
- Your Birthday, Overruled: ChatGPT can override your stated birthday and automatically activate the teen experience. Prediction works as an additional safety checkpoint.
- What It Watches: Signals include your topics, typical active hours, usage patterns over time, and account longevity. Behavior quietly stands in for birthdays.
- Check Your Label: Wondering whether ChatGPT quietly considers you a teenager? Open Settings, Account, then Age to see your currently assigned age classification.
- The Selfie Clause: Misclassified adults verify through Persona, providing a live selfie, government ID, or both. OpenAI insists it never accesses either submission.
Age checks are shifting from a checkbox you tick to a pattern you reveal. Expect more platforms to estimate maturity by behavior rather than paperwork. Your late-night chats now say something about you.
|
|
|
How to Hire an AI Employee Inside Slack:
Most AI tools are great at giving you information. They summarize, suggest, and explain. But when the meeting ends and the to-do list is still there, the actual work is still yours to do. That gap between “AI told me what to do” and “AI did it for me” is where most tools stop.
Viktor is an AI employee that lives in Slack and Microsoft Teams. You talk to it like a colleague, it connects to your tools, and it delivers actual outputs. Reports, dashboards, spreadsheets, code, campaigns. Not summaries of what you should do, but finished work.
Here’s how to get started:
- Install Viktor in Slack or Teams: Add Viktor from the Slack App Directory or Microsoft Teams in about two minutes. No complicated setup and no new interface to learn. Viktor lives exactly where your team already works.
- Connect your existing tools: Link Viktor to whatever you already use, Stripe, Notion, HubSpot, Google Ads, Meta Ads, GitHub, Linear, and over 3,200 others. Once connected, Viktor can pull data from all of them in a single message without you switching tabs or exporting CSVs.
- Give it tasks like you would a real colleague: Type something like “@Viktor audit our Meta Ads and Google Ads spend vs last month” or “@Viktor build me a Q1 profitability report” and it gets to work. It queries your tools, runs the analysis, and delivers the output directly in Slack.
- Get real deliverables, not just text: Viktor does not reply with bullet points telling you what to do next. It sends back the actual PDF, the spreadsheet, the deployed web app, or the pull request. The work is done when it responds.
- Set up recurring tasks and automations: You can schedule Viktor to deliver a business pulse every morning, assemble an investor update every month, or run outbound sequences weekly. It proposes automations on its own too, based on what it notices your team keeps doing manually.
If you are already using AI tools but still doing most of the actual work yourself, this is what that next step looks like.
|
|
|
- daily poll and result - |
|
|
Today's Poll:
| Q) Should AI agents be allowed to sign contracts digitally? |
|
|
|
Vote and find out the result tomorrow.
Yesterday's Result:
Q) Will AI shift power from labor to capital?
- A) Yes, automation favors owners - 98% 👑
- B) No, new roles emerge - 2%
|
|
|
Arcwise AI: Analyze and transform spreadsheet data using natural language directly inside Google Sheets.
Canny AI Autopilot: Automatically categorize, summarize, and extract insights from customer feedback using AI.
WhisperTranscribe: Turn long audio files into structured blog posts, show notes, and transcripts automatically.
Durable AI: Instantly generate a complete business website with AI copy, images, and layout.
Hotpot AI: Create graphics, images, and marketing visuals using AI-powered design tools.
|
|
|
The Colossal Titan Peers Over Shiganshina:
Prompt:
"Semi-realistic anime key art of the 60-meter Colossal Titan peering over Shiganshina's intact stone wall, gripping the battlements. Exposed red musculature, detailed muscle fibers, and thick superheated steam rising into a cloudy late-afternoon sky. Tiny human silhouettes stand atop the weathered masonry, looking up. Below, a dense medieval town with tiled roofs and chimney smoke. Terrifying, monumental, awe-inspiring scale."
|
|
|