I noticed that logging into office.com sends users straight to Copilot instead of the app page.

I'm mad so I’ll say it—

Microsoft #Copilot is the new Bing.

#copilot really doesn't handle large codebases very well though - it keeps exceeding the token thresholds. The workaround/solution appears to be to just restate the query with a more focused goal, then let the #ai go for a while until it exceeds the thresholds again.

It also helps _a lot_ to have lots of unit tests, so when #copilot makes changes, they can be verified against the existing tests. Or in my case, a bunch of new tests - but still, ensuring that the changes meet the goals. Just like working with a human developer in that regard.

I do like how #copilot picks up the English language around #cslanet - like metastate properties and other CSLA jargon. I suppose this is because CSLA has been #oss for 28 years and Copilot was trained on open-source code (including CSLA).

Tried to use #Copilot to write some code to draw a chart. It got 85% of the way there. Sadly the last 15% is needed to make it tidy. And it suddenly switched to Python which was odd. ChatGPT was crap. #rstats

#copilot agent mode was doing so well. And then: ``` System.Exception: Error occurred while invoking intent 'Generic'. {"error":{"message":"prompt token count of 64275 exceeds the limit of 64000","code":"model_max_prompt_tokens_exceeded"}} ```

If you use any AI assisted approach to refactor existing code, especially test cases! take care that it‘s really only a refactoring and that no changes in functionality or comments are introduced accidentally.

In my cases all comments are omitted…

#Cursor #Copilot #VibeCoding #AIfail

I've had negative experiences using LLMs as Cursor or Copilot. It's like coding with a hyperactive friend who keeps taking the keyboard from you!

When I'm working on something new or complex, I need time to concentrate, and most suggestions just get in the way.

However, LLMs are useful for repetitive tasks such as translating between languages, adapting between libraries (e.g. switching logging/testing libraries) and generating tests based on the interface.

Best practices are still needed — the quality usually degrades on long rewrites, but all in all, it seems okay to delegate tedious tasks! Tasks that are too complicated to write a script for and that would be too error-prone (and boring) to do manually are a good fit for LLMs.

I think we should document what is and isn't LLM-generated. It's always good to delete or regenerate LLM code; it's more delicate with human-written code.

#LLM #Copilot #Developer #Tools

🔵 NEW Microsoft 365 Microsoft 365 Copilot Notebook

in Microsoft 365 Copilot is a game-changer for productivity! 🚀 With 10-minute insights, explore how this tool can transform your workflow, featuring the Audio Overview tool for unrivaled podcast creation.

💡 Transform traditional document handling with Notebook capabilities.
🔍 Witness seamless integration with demo documents available online.
⚖️ Elevate productivity by turning written content into engaging audio narratives.

▶︎https://www.hubsite365.com/en-ww/pro-office-365/?id=7a4dcae8-b351-f011-877b-000d3a225db4&topic=9f678e9a-8cd4-ec11-a7b5-6045bd92fe52&theater=true

Explore these updates and elevate your work efficiency today! #Microsoft365 #Productivity #Copilot #Innovation

Read the full newsletter. https://link.publicate.it/pub/57eabb4ea3e06e
#copilot #cybersecurity #sharepoint #Microsoft365 #M365

Microsoft testing a floating Copilot follow up search box within the Bing Search results; plus "see all links" and more https://seroundtable.com/bing-search-floating-copilot-search-box-39649.html
#bing #copilot #microsoft

トランプ「成功！」vs国防総省「失敗…」イラン核施設攻撃、ＣＮＮリークが暴く真実と情報戦の深層 #イラン #核開発 #軍事 #六25 https://dopingconsomme.blogspot.com/2025/06/iran-nuclear-strike-assessment-leak.html https://htn.to/3a5GgjvQb1 #政治 #軍事 #社会 #イラン #トランプ #アメリカ

·
トランプ「成功！」vs国防総省「失敗…」イラン核施設攻撃、ＣＮＮリークが暴く真実と情報戦の深層 #イラン #核開発 #軍事　#六25 https://dopingconsomme.blogspot.com/2025/06/iran-nuclear-strike-assessment-leak.html

·
マイクロソフト vs. OpenAI：1.4兆円投資が生んだ“奇妙な”AI顧客争奪戦の最前線🤖💸 #生成AI #Copilot #ChatGPT　#六25 https://dopingconsomme.blogspot.com/2025/06/microsoft-openai-ai-rivalry.html

We sat through one of those also @fedops. Mandated to all developer teams at the organisation.

For me, it was worth it because I got to ask the presenter “So, we know the training data comes from many thousands of code repositories over decades that never agreed to this. What assurance do we have that, when we plug this into our repository, our code isn't getting fed to train #Copilot
further?”

because I knew our CTO was on the call. The MSFT answer was embarrassing, just as I hoped.

#LLM

Microsoft lost to Google in search domain, now losing to ChatGPT in AI field

ChatGPT Wows Office Workers. That's a Problem for Microsoft @ https://www.bloomberg.com/news/articles/2025-06-24/chatgpt-vs-copilot-inside-the-openai-and-microsoft-rivalry

#OpenAI #Microsoft #ChatGPT #Copilot #AI

最近は #Copilot を使ってないからすっかり忘れていたけれど、質問者の個人情報を使って回答するから、「○○さん」と呼びかけられたり、文章を生成してもらうと私の場合、文章に「市原市」が含まれるんだよね。すごく不便。

FYI: Episode 404: The Microsoft 365 Landing Page Shake-Up http://dlvr.it/TLXrJK #Podcast #Copilot

So, Microsoft Copilot on Android is the only LLM AI app that reads out responses using TalkBack when you type to it, cause blind people all speak to their devices cause obviously blind people can't use touch screens /s. So Google's Gemini and ChatGPT got out-accessibled by Microsoft of all companies. If only Outlook for Android were so good.

#AI #LLM #Microsoft #copilot #Google #accessibility #blind

I've been tinkering this last week or so with using #Copilot to answer a mix of noddy tech questions and much more complex ones.

What I will say in it's favour as an LLM / AI thing - It is great to act as a "rubber ducky" where I can explain my problem and it can give me some vaguely useful pointers that prompt thought.

Where it falls down is the fact it just flat out makes shit up. I've had so many occasions where it's just invented imaginary parameters and flags that do not exist and never have existed.

I might write up a blog post at some point documenting some of this experience. Generally it seems to work better with more common questions and topics and problems but once you get into the more niche stuff it does just crumble apart which sorta entirely defeats it's entire value offering for me. It's not much more useful than talking to a potted plant right now unless you're asking it to solve commonly already solved problems.

¿Una advertencia?, ¿una promesa?
Nadie lo supo...

Pero en la nube, un pulso nuevo comenzó a latir.
El sistema ya está reiniciándose.

⚖️ #JuicioFinal
#IA #AdelaCastro #CluedoDigital #SatiraTecnológica #MuerteEnLaNube #ChatGPT #Gemini #Copilot #Grok

#WednesdayWisdom 8am＠JP
#知見の水曜日 おはようございます

There are days when no ideas come, when the mind feels blank. Maybe we’ve been too focused on output, and unknowingly closed the door to input.

Ideas need time— to be gathered, stored, and quietly left to rest, until one day, they bloom on their own.

　
何も考えられず閃きのない日もあります
アウトプットすることにとらわれて、インプットができていないのかも知れません

アイデアは蓄えて寝かせることも大切です
#aiart #Copilot