Note-taking-ai

A bumper fireside chat this week – stuffed a lot into a short time. And to top it all, my suggestion on recording the meeting, and then asking gemini to extract the transcript and summarise, did not prove as prosperous as I thought. To demonstrate, I will add it to the bottom of these notes. Maybe the other AIs would do a better job, but Fireflies seemed to knock the spots off the gemini attempt.

KEY TAKEAWAYS OVERVIEW

  • AI Note-Taking Criteria: Transcription accuracy (25%) and summary quality (20%) prioritized; South African accent support and data privacy key concerns.
  • Tool Comparisons: Fireflies popular but intrusive; Granola botless with good local accent support; Jamie AI offers strong free use; Copilot fits Microsoft 365 users.
  • Privacy Approach: Local meeting recording plus AI transcription protects data; scripts for this method to be shared for wider use.
  • AI Market Trends: ChatGPT 5.6 introduces three tiers; US government invests in Anthropic and OpenAI; privacy and ethical issues rising with AI content misuse.
  • User Insights: Preferences split on bots; botless tools favoured for authentic presence; Granola praised for local accent accuracy and live transcription.
  • Operational Needs: Compatibility with Teams/Zoom and mobile app availability vital; data retention policies and export options impact trust and compliance.
  • Local AI Solutions: Running AI models locally offers privacy despite slower speed; hardware support discussed to improve capacity for private processing.
  • Cultural Nuances: South African English accents challenge AI accuracy; meeting bots viewed variably as efficient or impersonal based on cultural attitudes.

Notes

AI Note-Taking Tools Evaluation

The team reviewed key criteria for selecting an AI note-taking app, focusing on transcription accuracy, privacy, and usability.

  • Core evaluation criteria for note-taking apps weighted transcription accuracy highest at 25%, followed by AI summary quality at 20%, with confidentiality and security at 15% (05:50)
    • Malcolm emphasised the importance of supporting South African accents for accuracy given local dialects
    • Privacy concerns include data residency preferably within South Africa and control over data retention and deletion proofs
    • Integration with meeting platforms like Microsoft Teams or Zoom affects tool suitability
    • Pricing models vary by user, minutes recorded, or credits, impacting cost-effectiveness
  • Comparison of popular AI note-takers highlighted Fireflies, Microsoft Copilot, Fathom, Granola, Otter AI, and Jamie AI (14:00)
    • Fireflies is popular for ease of use but uses bots, which some users find intrusive
    • Granola, preferred by D, offers botless operation with live transcription and good support for South African accents
    • Jamie AI provides a strong free tier without bot involvement, appealing for cost-sensitive users
    • Microsoft Copilot suits organizations heavily invested in Microsoft 365
  • Alternative approach using audio recording plus AI transcription services like Gemini or Claude was proposed to maintain data control (16:20)
    • Recording meetings locally then using AI tools for transcript, summary, and action extraction can replicate note-taker app functions
    • This method addresses privacy by keeping raw data in-house before processing
    • Malcolm plans to share scripts and guidance for this approach in meeting minutes
  • Bot attendance in meetings presents trade-offs between convenience and perceived engagement (08:50)
    • Bots can capture meetings outside working hours, helping users avoid inconvenient times
    • However, some attendees view bots as a sign of disinterest, potentially harming relationships
    • D prefers botless tools to maintain a more authentic meeting presence

AI Market and Industry Developments

Recent AI advances and market moves were discussed, showing rapid evolution and increasing government involvement.

  • ChatGPT 5.6 release marks a significant leap in AI capabilities beyond incremental improvements seen in prior versions (21:20)
    • The update includes three tiers: Seoul (advanced), Terra (balanced), and Luna (cost-efficient)
    • Industry observers suggest the jump is large enough to merit a version number 6 designation
  • Government investment in AI firms is increasing, with Anthropic and OpenAI both selling 5% stakes to the US government (26:50)
    • F noted this move supports data centre operations and reflects growing government influence in AI sector
    • D expressed concern over government stakes influencing AI access and data control
  • Emerging privacy and ethical issues around AI-generated content and social media were highlighted (18:50)
    • Instagram’s default setting allowing use of user photos for AI image generation raises consent concerns
    • A lawsuit alleges Grok AI created 7,000 abusive images from one child’s photo, underscoring risks
    • Academic responses to AI cheating include moving exams offline, with reported score drops from 96 to 48
  • Other notable AI-related news included The Economist launching an anti-AI campaign and environmental impacts of data centres (19:15)
    • Data centre wastewater contains rare bacteria, raising ecological concerns
    • Social media and AI may contribute to teen mental health penalties estimated at 1.4 trillion dollars

User Experiences and Preferences

Participants shared their personal use cases and preferences for AI note-taking tools based on practical needs.

  • DW prefers Granola for its botless design, ease of use, and accurate South African accent transcription (22:40)
    • He values the ability to pause recording until all attendees are present
    • Granola’s live transcription and meeting summaries help him quickly review content
    • Operates on Windows 11, showing cross-platform availability
  • Malcolm Pearson uses Fireflies but is cautious about bot involvement and underutilizes allocated credits (14:00)
    • He is exploring alternatives like Jamie AI and Microsoft Copilot depending on organizational fit
    • Plans to circulate a detailed comparison table to help others make informed choices
  • Concerns about balancing AI tool convenience with meeting culture and engagement were discussed (09:00)
    • Bots reduce personal attendance but may send unintended signals about commitment
    • Users weigh trade-offs between automation and authentic presence

Operational and Technical Considerations

The group examined technical and process aspects influencing AI note-taking adoption and implementation.

  • Meeting platform compatibility is critical for AI note-taking tool selection (07:50)
    • Some apps work better with Zoom, others with Microsoft Teams; Copilot is Teams-centric
    • This affects workflow integration and user adoption ease
  • Mobile client availability for iOS and Android is essential to support on-the-go note-taking (08:10)
    • Participants stressed the need for recording and transcript access outside the office
    • Apps with mobile versions enhance flexibility for hybrid work environments
  • Data retention policies and export capabilities are important for compliance and data control (07:40)
    • Users want clear assurances that data is deleted after export and not used for training AI models
    • This impacts trust and willingness to adopt cloud-based services
  • Recording meetings locally using built-in OS apps or external recorders offers data sovereignty (16:00)
    • Using local audio files with AI transcription tools mitigates cloud privacy issues
    • Malcolm plans to share scripts for extracting transcripts and action items from these files

Future Infrastructure and AI Tool Development

The discussion touched on resource sharing, system setups, and evolving AI tool usage within the team.

  • D is running LM Studio locally for confidential AI processing despite hardware limits (28:40)
    • His laptop supports about 8 billion instructions, sufficient for private tasks but slower than cloud AI
    • Malcolm offered to send spare PC hardware to D to improve capacity and assist with setup (29:20)
  • J is also experimenting with personal LLM setups, signalling a trend toward decentralized AI use (28:50)
    • The group sees value in running independent AI models to maintain data privacy and reduce external dependencies
  • Time constraints limit full adoption and experimentation with AI tools (30:00)
    • Malcolm noted that side projects excite but compete with income-generating work
    • The team recognizes balancing enthusiasm with productivity challenges
  • Sharing discoveries and best practices about AI note-taking apps is encouraged to enhance collective knowledge (31:00)
    • Malcolm invited participants to share their best finds so everyone benefits
    • This aligns with a culture of continuous learning and practical improvement

Cultural and Linguistic Nuances

The meeting highlighted challenges and humor around local language and terminology affecting AI and communication.

  • South African English accents pose challenges for AI transcription accuracy (06:00)
    • Malcolm recounted historical struggles with Dragon Dictate and current improvements
    • D confirmed Granola’s good handling of local accents despite occasional misnamed companies
  • Misunderstandings around local legal and land registry terms provided light relief (25:30)
    • Malcolm shared a story about confusing “land tenure” and “deeds registry” terminology from official announcements
    • This underscores the importance of clear communication in public sector messaging
  • Attitudes toward bots in meetings reflect cultural preferences for personal presence and respect (09:00)
    • Some see bots as efficient, others as impersonal or dismissive
    • These perceptions influence tool acceptance and meeting etiquette adaptations

Action items

Malcolm Pearson

Attach the AI note-taking applications comparison table to the meeting minutes and distribute it to the group (15:15) – Attached below.

Ship a spare PC to DW for installation and testing of LM Studio AI software (29:20) Fireflies not getting the humour here – don’t stress DW….

Attach link to strange new products and words for Deeds Registration EDRS announcement on the topic of strange accents .
https://www.linkedin.com/posts/carlize-knoesen-51779055_erdrs-amendment-bill-activity-7478784110111440896-ST51?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAAYqpwBga-V4g8HlzgF93uk1tg5gi70KOE

DW

Install LM Studio on the spare PC received from Malcolm and return it after setup and testing (29:30) – Not going to happen – you can relax DW 😉

All Participants

  • Evaluate AI note-taking tools based on criteria such as transcription accuracy, privacy, integration, and bot presence, and share feedback and best findings with the group for collective benefit (30:50)

The PDF

~

The transcript and summary by gemini…

Transcript: AI Collaboration Discussion

Date: Thursday, July 16, 2026

Participants:

  • Malcolm: 0:01
  • K: 0:02

Transcript:

Malcolm: K, good afternoon. How’s it going, my area?

K: Hey, 100%, how are you?

Malcolm: Good, man. Good. How’s the weather, there, hot?

K: Yeah, no, it’s—it’s actually been pleasant, hey. I went out at lunchtime and it was like—it was nice and warm. Yeah, 24, 25 degrees, so pretty cool.

Malcolm: But, but after the sun goes down, it gets a bit nippy.

K: Very, [laughs] very different picture, yeah. Yeah. Yeah, it is.

Malcolm: 100%. Uh—let me just close this down. Um—Are you at work or are you on holiday?

K: No, I’m at work, hey. Yeah. I’m actually—half-day tomorrow, and Friday I’m off, so yeah, looking forward to a long weekend.

Malcolm: Okay, nice. Are you—Are you going away?

K: No, no, no, I’ve just got a bit of work to do and my wife’s on holiday as well, so she’s—she’s at, uh—at school, so she finishes at the end of the week, so I just catch up the last few days with her and then, yeah, she’s back to work as well.

Malcolm: Okay, nice. Cool.

Malcolm: Uh—yeah, I’m just waiting for a couple other people to come on board and then we can roll in.

Malcolm: Um—is your rollout stall, uh—stall on—on hold?

K: Yeah, it is at this stage, hey. Yeah. Yeah.

Malcolm: Yeah, let me, let me. Hopefully, hopefully, get it resolved eventually. Um, it’s all on our side, not on their side, so, yeah.

K: Yeah, okay.

Malcolm: The delay.

K: We hope—hope and pray that it happens. That’ll be good.

Malcolm: Yeah. Yeah, yeah, it’s—it’s a—it’s a big change, but, uh, I suppose once you’ve gone through it, it’s fine, it’s just getting through it.

K: Yeah, that’s it. Yeah, yeah. Yeah.

Malcolm: 100%.

Malcolm: [Coughs] So, uh, at once did you—were you guys ever on R governed to stuff, uh, Law—Law—?

K: Pack, was it? We’ll call it that.

Malcolm: Low pack was it?

K: Yeah. Yeah, we use Law Pack. Yeah.

Malcolm: Law Pack was it? Yeah.

K: Yeah, that is Law Pack. Yeah. No, I—I don’t. That was yeah. I mean, we’ve been on on legal speed for like 20 years plus, so we’ve been on it for a very long time, yeah, so

Malcolm: Okay.

K: Um—yeah.

Malcolm: That’s—been a while.

K: Yeah.

Malcolm: Yeah, yeah, R, R always used to tell the story of how he did work for people and then they would say, no, we’re going to another provider. And then R would say, well, we’re very sorry to see you go and everything. And then they would—he’d help them across, and then about six months to 8 months later, they’d come back and say, hey R, we want you to enable our stuff again, and we want to come back to you, and he—he used to—he used to double or triple the price. And then they’d say, no, we still interested, and then they’d take them back. [laughs]

K: Yeah. It’s often that—a—a, it’s a funny thing, hey. That’s a very funny thing.

Malcolm: Yeah. People get used to it.

K: Yeah, absolutely, hey.

Malcolm: Yeah, okay. Thanks, K.

K: All right.

Malcolm: Right, everybody—or K, D, and Malcolm, it’s good to see you guys there. But, uh, yeah, so, the the notetaker AI notetakers that are available, I wanted to chat about that a bit today. Um, I know, D, you’ve mentioned before, you use one of the others and, when I say one of the others, I’m talking about, um, Fireflies, but, um, yeah, so, I wanted to sort of cover those, uh, as a—as a basic overview of things you can look out for and what you should evaluate when looking at them.

Malcolm: Um, so, yeah, and to—to kick off, I’ve sort of gone through and made some notes, so I’ll just follow those notes. Um, on the AI notetakers, um, before deciding on a system, what you should look at is accuracy. Now, I will talk about these things in a little bit more detail, but accuracy, privacy, where the data is going to be stored, where most of your meetings are being hosted, uh, whether you want a bot or not, a bot or not—sounds like an anibotobotomy, but—a lobotomy, uh, price, and then mobile client, and then searchability of the data that you have in there.

Malcolm: So, speaking of accuracy, and this will sort of make you chuckle, but us South Africans have a lack of kiff way of tuning each other. So, you can imagine, if you have people talking like we do, the South Africa, with our accent, uh, accuracy is always something you need to look at because then, I mean, up until, in the days gone by, Dragon Dictate always used to battle with South African accents, and it’s—you can understand why, I mean the rest of the world, the size of the markets they have, are absolutely monstrous and then they got South African with a couple of hundred thousand people there that want to use some sort of notetaking app, and our accent was always something that was in the last thing that they would be interested in fixing.

Malcolm: But, uh, I think it’s obviously, it’s a lot better these days, but you should look for—you’re talking of accuracy, make sure that the notetaking app that you’re looking at, make sure that it can actually handle our South—South African accent.

Malcolm: Then on privacy, um, ideally, you know, you want your data to be housed on your network, or on your PC. That’s first prize. Um, if it’s cloud-hosted, hopefully it’s inside the South African borders, if it can be, that’s second prize. Uh, third prize, etc., will be hosted outside of South African borders, but obviously, under strict control where you can control what’s happening to that data. Um, things you should ask those people then is, uh, you know, what is, what is your, how are you handling, uh, data retention of data? Are you using our data to train your LLMs? Um, and then also, if we, if we, if we export all our data from this notetaking app for what we’ve been using it for, um, is there a way to export, and once we’ve exported our data, is there a way of us getting proof that it actually has been deleted and not being used?

Malcolm: So, that’s on data stuff. Then, where are your online meetings being held? That’s sort of important, depending on the client that you’re looking at. Some clients only work with, uh, with certain meeting types, like, for example, a very good example, is Microsoft, uh, Teams and then, obviously, Copilot would obviously be a choice for you looking as a note-taking app, but it would obviously be, it only works with Teams, uh, or works most, it works best with Teams. Um, obviously, with, uh, working with the meeting meeting types, some work better than others, so you’ll find that one it works well with with Zoom, but doesn’t work well with, with Teams, etc., so you need to look at that.

Malcolm: The mobile client, what happens when you are out of the office and you have a want to, want to record something or record a meeting, uh, and take the minutes of that meeting, or get notes or transcript, that you need to take into consideration. You obviously got iOS or you got Android, etc. Those are the basically the two these days. Um, and then, bot or not, lobotomy, as I said earlier, the bot or not, so, the bots are like in Fireflies, the nice thing about a bot, like an agent type bot, is at 10 o’clock on a Friday evening, uh, the last thing you want to do is attend a, uh, a webinar that’s happening in New York, and to them, it’s fine because it’s still sort of halfway through their day.

Malcolm: But for us, it’s 10 o’clock at night, you—you couple you—you’re obviously a couple of drinks into your festivities of the evening and gather around a braai, the last thing you want to do is excuse yourself and go and attend a meeting. So, those are nice because you get your desk on a Saturday morning, and before you can wipe that, uh, sleep out of your eyes, you can see the actual transcript or the meeting notes from the previous night. So, so that’s nice. The downside of that is people look at bots attending meetings as, um, as a message being sent that you’re not really interested in the, in what they’ve got to say because if you were interested in attending the mi—you know, in what they had to say, you’d be attending the meeting yourself, in person.

Malcolm: Um, so it—it lacks a bit on that side, and some people actually have a problem, you know, of, of bots being part of a actual meeting. Um, so there’s that side to it. Uh, then searchability, down the line, you know, you want to be able to check on something like you might remember, oh shit, in August last, last year, um, you know, did, did, in that meeting that we were chatting to Fred, did he admit, did he actually admit to the fraud in that actual meeting? And then you want to be able to go and search your trans—so your transcripts, to make sure that he actually said that. And actually, if he did, then you obviously have it. But make sure you can search through the system and the data that you’ve collected. That’s why it’s there, really. And then pricing, uh, pricing varies, some have per month per user, uh, others are by the minutes of the recordings of the meetings, um, the others are by credits, so, where you—when you’re attending a meeting, it’ll obviously allow so many credits, or take up so many credits, uh, to actual record that meeting.

Malcolm: Um, so, that’s on pricing. So, so how I would score these, these various aspects, uh, criteria basically you could say, is, uh, transcription accuracy—we spoke about earlier, that’s got to be the most important thing. So, we sort of weighted at 25%. The AI summary quality, we’re looking at about 20%, needs to obviously, if it summarizes, you need to be accurate. Um, confidentiality and security, of your data, 15%, ease of use is 10%, search and retrieval is another 10%, integrations to other systems, um, that is at 5%, action item extraction, you know, what actions need to be taken, 5%, export options, another 5%, and then pricing and value, sort of 5%, and I would imagine depending on budgets, yeah, that whether that’s going to be 5% or not, but what it breaks down to is that you basically, the most important thing is that your the quality of the actual transcript is actually most important for everything.

Malcolm: Um, right, and then the note-taking apps that I looked at, uh, that have of, well, that sort of come to mind when you look at note-taking apps, Fireflies is the one that I’m currently using. Uh, and I must say I’m starting to look around at others. Uh, the bot is a worrying thing. Uh, the other thing is I’m not using it for all the credits that I actually have, I mean, I don’t use it, uh, a hell of a lot. Um, Copilot is an option, and it’s a good option if you inside of Microsoft, Fathom is a good one. Uh, Granola is another one that’s gaining momentum, Otter AI has been around for a while, and then Jamie AI, and then Jamie AI is a bot-less version, um, but what I can do is I’ll share my screen, but let me show you, uh, this assistant, this, uh, notetaking, um, sort of table that shows you the various ones. Let me just share my screen.

Malcolm: There we go.

Malcolm: Can you guys see my screen?

K: Yep, can see it.

Malcolm: Excellent, thanks Malcolm. Um, okay, so, so there it sort of goes through the five, you know, it doesn’t mean that they are in that order, that’s the way they should be, that’s not first, second and third, that’s just going through the various ones and what they offer. Um, but the Fireflies seems to be very popular, uh, it does work quite easily and seamlessly, works well with most of the stuff, um, Microsoft Copilot, if you are a Microsoft shop, and have Office 365, with the Copilot license, then, uh, that’s probably a good way to go. Um, and then Fathom, I don’t know that much about Fathom, but, uh, also, sort of a bot version, um, Granola doesn’t have a bot, so that’s depending how you, um, how you, you know, whether you’re for a bot or not, for a bot, and then you see that strange thing there, they say that the audio is not retained, which is quite interesting, and I don’t know whether they’re talking about obviously, on, in the cloud, it’s obviously not retained, you’re obviously saving it to your local machine.

Malcolm: Um, and then Otter also Otter, also has a, um, a bot, and then Jamie AI seems to be people that are talking quite a bit about it, um, and it says it seems to have quite a, a very good, um, free option, so they say that the free option quite offers quite a bit before you actually have to pay for something. And Fireflies does offer you to record one or two meetings, uh, on the free version. Uh, I think it—I think it’s three in a month, that it allows you to actually record and then use and then you need to pay for it after that. But this, uh, comparison what I’ll do is I will, um, okay. I will attach it to the end of the, the minutes that I send through, uh, tomorrow. Um, so you can actually see that.

Malcolm: Um, yeah, so that was the comparison table and how the various ones look. Something, something that I wanted to share with you outside of that, so that’s basically the apps that are available or some of the apps that are available. They are whole plethory of them, that you can actually go through. You know, obviously it depends what fits all the things that you needed to, to do. Um, some things to ponder though, is I was thinking if you take a, a recorded meeting—so we use Zoom, and, uh, you can record Zoom, so you can say you can record the audio, which this one is doing. So it records the audio in the background and then it saves it to your local machine. So basically, you have an, an audio file that’s available to you of the whole meeting, and if you don’t have, if it doesn’t happen inside your meeting, uh, your online meeting system, you can use something like Recorder, um, on the, the Mac, it’s called, uh, Voice Memo or QuickTime and those are standard apps on there. And in Windows, there’s something called Sound Recorder in the newer versions, or Voice Recorder in the older versions. Um, or, or on your phone, you can use a recorder, something, if you haven’t got a recorder, you can use something like Philips recorder, which is actually quite good.

Malcolm: Um, but those are all produce the audio file for you. So, so basically, if you, you want to keep the data to yourself, you record your meeting audio, and then you go into Gemini, and Gemini, I, I, went into Gemini and I asked them about the transcripts and stuff, and they said, yeah, well, you can go and, you know, you can ask it to do this, so the script that it came up with, and I’ll include this in the minutes as well, please provide me with a verbatim transcript of this audio file, including speaker labels and timestamps. Following the transcript, provide a concise summary of the main points discussed and a bulleted list of all action items identified in the meeting. That, now, there, it does everything that, uh, I would want, uh, Fireflies to do for me, there, Gemini does it and it’s there’s not, there’s not the, the, the best, the greatest thing, uh, I’m using the free version there, but that’s something that can do for you and it does everything that Fireflies would do anyway for me.

Malcolm: So, I like that. Uh, that fact is that you’re doing, the audio is a recording, so you’re getting that anyway. Then you’re getting the transcript, and the whole minutes and the summaries and the action points, all from Gemini, which is doing it and I suppose, once we talk about data, and we go and dump this thing into Gemini, I suppose then our data all goes all over the place. But let’s say you’re using Claude Desktop and it’s private, you can actually use Claude to do the same thing. I’m sure it’ll work just as well. But what I’m saying is, you know, you, you could get the same thing that you’re getting from a, uh, from a meeting notetaking system, you can get from just recording the audio, and from actually using something like Gemini.

Malcolm: So, that I thought was something that people don’t really think about. I mean, I, I just thought about that because I thought well people are using AI and the AI systems can do this anyways. So why don’t we use them to do that, anyway? It makes sense. Then, um, J gave me a heads up earlier, earlier this week, I think it was. Um, so he he got a, a notification from one of the, the forums that he follows, and it was quite something, actually. So they were talking about how what had happened in AI in the last 48 hours, and it was quite, uh, eye-watering to see. So Instagram now lets people use your photos to generate, generate AI images, and it’s enabled by default. So obviously, you can go and switch that off, but that means that somebody can go and use your images, uh, you know, and generate, generate AI images using your images that you have in in Instagram. That’s quite something. You don’t really want other people creating stuff of you without your info, without your knowledge.

Malcolm: Um, and then a lawsuit alleges that Grok generated 7,000 abuse images from one child’s photograph. That’s also not very nice. I mean, I suppose these things are flying around all the time. Um, and so, yeah, he has a a chuckle, a professor suspected that the students were using AI’s to cheat on tests. So what he did is he moved the exam offline and he says, scores dropped from 96 down to 48. So that’s, I suppose, it’s happening in all tests. Uh, the the economist just launched an anti-AI campaign, uh, and then bacterium, a rare bacterium found in wastewater tied to metadata center, you know, these big data centers and obviously they have to have cooling and the water they’re using through the cooling, and now it’s putting bacteria back into the water. Not a very nice thing if their water is going further down the, down the line. Um, and then it says that Meta is facing a 1.4 trillion Rand in penalties in teen mental health care. Um, yeah, so they’re saying that teens are um, mental health with obviously what they’re doing with AI and social media etc.

Malcolm: But, the other big thing that happened, uh, it’s probably a little bit longer than a week ago, um, Chat GPT launched fire version 5.6 in the versions, three versions of it, in 5.6. Uh, there was Soul, which is the most advanced user level, then Terra, the balanced, and then Luna for the most cost efficient, obviously the cheapest and the quickest and the nastiest. Um, but people are raving about Chat GPT 5.6, but then they’re also saying that Meta’s also released something and somebody else has released something, um, but they believe that that grounds—well, you know, we’ve been ticking along for a while now where people have put new versions out of the LLMs and it’s been nice and it’s it’s been sort of incremental. The one or two features come out, one or two features and it gets slightly better and slightly better. But they say with this Chat GPT 5.6, it’s that much of a jump that they, they feel that people that were reporting that it should have jumped to like a Chat GPT version 6.

Summary of Main Points

  • AI Notetakers: Malcolm provided an overview of factors to consider when selecting an AI notetaker, including accuracy (specifically handling South African accents), data privacy/hosting, integration with meeting platforms, the benefits and drawbacks of using “bot” agents, searchability of archives, and pricing structures.
  • Alternative Notetaking Method: Malcolm suggested a privacy-conscious alternative where users record meeting audio locally and then use an LLM (like Gemini or Claude) to generate a verbatim transcript, summary, and action item list.
  • Recent AI Industry Developments: Malcolm shared a series of recent industry updates reported by J, including:
    • Instagram’s default policy enabling user photos for AI image generation.
    • A lawsuit against Grok regarding the misuse of a child’s photo.
    • Academic observations of significant score drops in exams after moving them offline to prevent AI cheating.
    • The Economist launching an anti-AI campaign.
    • Environmental concerns regarding bacteria in wastewater from data center cooling systems.
    • Significant legal challenges facing Meta concerning teen mental health.
    • The release of ChatGPT version 5.6, which industry observers characterize as a major leap forward rather than an incremental update.

Action Items

  • Malcolm: Attach the AI notetaker comparison table mentioned in the meeting to the minutes.
  • Malcolm: Include the specific prompt script suggested for generating transcripts and summaries via Gemini in the meeting minutes.
  • Malcolm: Add the comparison table for various AI notetaking apps to the minutes.