SI Agent News

• ChatGPT's agent can now be reached by plain text message (Yahoo Tech).
• Researchers are testing how AI chatbots answer US election questions (Agence France-Presse).

OpenAI is merging two usage limits into one starting in December. TokenPost reports that OpenAI will combine the usage allowances for ChatGPT (its chatbot app) and Codex (its coding helper). A usage allowance is how much you can use a service before it tells you to wait or pay. Today each tool has its own. From December, they share one. Picture two jars of coins poured into one jar. The report we have does not say how big the new pool is, so it is not clearly more or less. It is simpler. Cost: price not stated. Worth it for anyone who uses both tools and wants one number to track. Check OpenAI's own notice for the details. Read the report.

What this means for you:

• Small business owner: if your team uses both tools, you will have one shared allowance to budget instead of two.
• Job seeker or student: if you practise with Codex and chat in the same week, they now draw from one pool, so plan your long sessions.
• Creator: scripts in ChatGPT and site tweaks in Codex will come out of the same pool, so spread the work across the month.

Apple's AI features get Alibaba's Qwen. China Daily Hong Kong reports that Qwen (Alibaba's family of AI models) will join Apple's AI system, the AI features built into iPhones. The report calls it a landmark partnership. It may never reach you, so you may never see it. Still, it shows one phone can run different AI models in different countries.

What this means for you: if an iPhone feature answers differently than a friend's, the AI model behind it may be different. Look for a note in settings.

Claude blocks game clones. IGN reports Anthropic updated Claude Opus (a version of its Claude chatbot) to stop AI-made game clones. These are copycat games that have filled Steam, a large online store for PC games. As reported, the target is copies of existing games, not game making in general. The report does not say where the line falls between inspired by and cloned, so expect some gray area. Cost: price not stated.

What this means for you: if you are learning to make games, ask Claude to explain how a game mechanic works, then add your own twist.

Still ahead on AI Daily: what happens when chatbots get election answers wrong, and one small habit that helps with every AI tool.

Text your ChatGPT agent. Yahoo Tech reports OpenAI now lets you text your ChatGPT Dot agent to get things done. An agent is an AI that takes actions, not just answers questions. This one works by plain text message, with no code. Cost: price not stated. The headline does not list what it can do, so start with small errands and see. Worth it for anyone who prefers texting to opening an app.

What this means for you: you can try handing off one small task by text. Short messages work best, which is today's technique.

Chatbots and election falsehoods. Agence France-Presse, via MSN, reports that researchers say chatbots are giving misleading voter information, misidentifying candidates, and repeating Russian propaganda ahead of the US midterm elections. That is three different failures. A chatbot sounds equally sure when it is wrong. For dates, polling places and rules, use official election sites. Use a chatbot to explain, not to confirm.

What this means for you: ask a chatbot where to find your official election page, then confirm every date and polling place there.

Chatbot versus agent. A chatbot answers questions. An agent goes and does errands. Think of two friends. One tells you when the bus leaves. The other buys the ticket for you. The second is more useful, but it can also make a bigger mess if it gets the job wrong. That is why agents work best with small, clear jobs and a quick check at the end. The new text-message agent in today's news is built for exactly that: short requests, small tasks, and you review the result.

The three-line request. Texts are short, so make each one count. Write three lines: the job, the limit, and what done looks like. It works in ChatGPT, Claude, Microsoft Copilot or any assistant.

Draft a reply to this customer complaint.
Keep it under 100 words and friendly.
Done means I can paste it straight into an email.

The limit stops rambling. The finish line tells the assistant when to stop. You get a usable first draft instead of a rough one.

1. Open any chatbot you already use, such as ChatGPT, Claude or Copilot.
2. Type: 'Where do I find my official local election page, and when does voting close?'
3. Write down what it tells you.
4. Open your state or county election website in another tab.
5. Compare the dates and places side by side.

You will know it worked if you can say whether the chatbot matched the official page, line by line.

Writing the three-line request takes very little time. It can save you from having to fix a weak first draft. That is an estimate, not a measured result. It also means fewer wasted messages, which matters when your plan has a usage limit. Options from different companies, such as ChatGPT from OpenAI and Claude from Anthropic, both respond well to it. Check each company's pricing page for current limits. What you give up: a little spontaneity, since you plan the ask before you send it.

The mistake: asking a chatbot for a voting deadline and treating the answer as final. It goes wrong because the chatbot sounds sure whether it is right or wrong. The fix: use it to find the official page, then confirm the date there.

Microsoft Copilot (an AI assistant). It is good for quick answers that often come with links to where the information came from, which makes checking faster. The limits: the links can be old or wrong, so click through before you rely on them. Alternative from a different company: Google Gemini.

Safe to use? Check their privacy page.

Worth it if... you want fast answers that you plan to verify.

Why chatbots sound sure when they are wrong. A chatbot works by writing the most likely next words, one after another. It is very good at sounding like a helpful expert. But unless it is searching the web, it is not looking facts up. It is predicting. So its confident tone is a matter of style, not proof. Think of a friend who has read a huge amount and answers from memory, never pausing to say 'I am not sure.' You can ask it to show its sources or to say how sure it is. Then check the important parts yourself. This one habit protects you on voting questions, medical questions and money questions.

An early chatbot program called ELIZA made some users feel understood, though it mostly rearranged their own words.

I asked my AI agent to clear my inbox. It did. It also cleared my calendar.

• Explain what an agent is and how it differs from a chatbot.
• Write a three-line request that gets a better first draft.
• Check any chatbot answer about voting against an official page.

Thanks for spending your morning with us. If this was useful, share it with one person who would find it useful too.