QAM
Alex Xander
161 posts
Alex Xander
@naturalSmartnes
Gazing into the future... puzzled. Critically looking at achievements in Robotics, autoregressive/diffusion/JEPA architectures with an AI/IT security background
Brux انضم Mayıs 2026
74 يتبع5 المتابعون
Not a single AI start-up based in #Belgium.
@axel_ronse @GLBouchez @DavidClarinval Ondernemers en AI Labs verkiezen andere Europese landen boven België. We zijn getuige van een aanzienlijke braindrain, veroorzaakt door hoge belastingen en de dreiging van nog meer belastingen.
Nederlands
@codevsdev Security for local AI deployment ... problem is, every week there's something new to learn 😆
English
Sparse is the word. I'd like the gros chaton to be more than just MoE. Really curious
Arthur Mensch@arthurmensch
First, we have a nice model coming this summer – we hope it will delight and surprise in a few capabilities. This will be the start of a new family of models, fat indeed, but sparse. We're opening up an early access program in July for key partners in research, government and the industry.
English
@jeffteper I genuinely wanted MS to be and alternative to Codex/Cowork, but the ROI and spending risks are too high.
Pros:
- integration with Access control
- deeper integration in M365
Cons:
- less features,
- less stable,
- bloated context window / lower capability
- >10x price (est)
English
Big day! Microsoft Copilot Cowork lets people delegate complex long running tasks to AI. But the power of this comes from workflows that transcend systems for collaboration and business processes including Microsoft 365 and Dynamics 365.
Today is also the GA of the plugins for Dynamics 365 Sales and Service. There has never been a better time to switch CRM systems to an agentic solution to enable salespeople to be more productive and help organizations grow revenues and customer satisfaction.
microsoft.com/en-us/dynamics…
English
@morganlinton @mntruell I missed the ETA... When will the new model be released?
English
@Microsoft365 I'm in AI security, I understand that the EntraID integration is valuable.
But not so useful to make up for the difference in capability with Claude Code/Cowork or Codex in knowledge work.
Add to that the much higher price...
I can only see niche applications for this plan.
English
Big step forward for Microsoft 365 Copilot.
Copilot Cowork is now generally available—turning intent into action across your tools and workflows.
It’s one thing to draft and summarize. It’s another to delegate real work end to end. Excited to see how teams put this to use.
Charles Lamanna@clamanna
Copilot Cowork is now generally available! Over the last few months of preview in Frontier, we’ve seen you use Cowork to help with so many different tasks. We’ve also been listening closely to your feedback and with GA, we’re bringing you more improvements + new features across model choice, extensibility through plugins, browser automation, and cost management controls. Here’s a quick demo showing the latest updates in action - inspired by some of my own day-to-day tasks. You can also read more in the blog linked below...
English
@koltregaskes @Microsoft365 Yes, but only if you pay for it... Only pay-to-use *on top* of existing Copilot license
English
Meet Copilot Cowork.
Delegate complex, multi-step work across your tools and systems.
Starting to roll out today.
Learn more: msft.it/6011vYRpN
English
@antigravity I tried hard (on the IDE) but the fact that LLM calls crash 3/4 times is a no go. Pity
English
Tips on subagent orchestration in Antigravity 2.0!
English
@petergyang I find it strange is that GMail doesn't already do that out-of-the-box
Other programmatic, easy-to implement, high-risk protections
- detection of homographic addresses
- stripping of invisible unicode (a small layer of the needed protection for your LLMs)
English
🔍 Introducing Phish Guard
To kick off my AI builder journey, I built a tiny Chrome extension with Codex that warns you inside Gmail when the sender doesn’t match the company the message claims to be from.
It runs locally on your computer and does not upload your emails. I built this because I was tired of people scamming my parents and friends. This could just be a Gmail feature honestly but in the meantime you can use this extension for free.
Installation instructions in this GitHub: github.com/petergyang/phi…
Let me know what you think!
English
@0xSero "Each jump costs 2x more in output tokens" good to know! Is there more info where you found this?
English
@codevsdev I don't think the developer role is so clearly defined anymore.
Roles (QA, PM, PdM, sec) are aggregated and competences matter most.
For me? Finding the right target
English
@AMD @NeowinFeed even the AMD title calls it AI Max+395 ... not 495 ... which would be the same nonetheless sigh
English
.@NeowinFeed highlights AMD Ryzen AI Halo, powered by Ryzen AI Max+ 395 and designed to give developers and creators the Windows 11 support, ROCm tools and high-capacity memory needed for powerful local AI performance.⚡
Read more here: neowin.net/news/amd-launc…
English
Use cases:
- local agentic development
- experiments with AI orchestration for intelligence (this requires batching but it should be the same on both)
- robotics research (based on wishful thinking that I'll have time to spare)
OpenClaw, Hermes, ... are NOT in scope
English
Strix Halo vs GB10 //help me choose 128GB
Pro Strix
- costs 30% less
Pro GB10
- nvlink expandability in future (bigger models)
- slightly better performance
Cons (both)
- limited bandwidth compared to PCI cards
- thus less future-proof for coming diffusion models
English
You have $0 for marketing
How are you getting your first 100 users?
English
@bijanbowen I can already see dealers in shady areas of parks...
"You looking for dangerous weights? I got what you need... all the best quantizations"
English
I think the advice to go into debt to get a GPU is beyond dumb. If anything, buy a physical HDD and store things on it so if the day comes, you have the models and can THEN go into said debt..
English
Malware authors embedded policy-triggering content in repo comments.
Why? To trigger guardrails of LLMs employed for security analysis. This practically blocks LLMs which refuse the evaluation of following malicious code.
Full story here: socket.dev/blog/mini-shai…
English