Allen

retrieval is lexically biased even when embeddings are good. ^^

@dotnet_pickaxe sure but if you have a tenant scale topping your available hardware then, unless you have a table ref tenant arch then all you have to do is move that tenant to its own larger machine, charge them more and you are good. db as tenant = best. hori scale stateless api layer.

I have an important warning for developers who are developing projects with #dotnet Multi-Tenancy: Isolation at the level of hardware and isolation at the level of isolation are not the same thing. As a result, we encounter the infamous "noisy neighbor" problem, which the sources particularly emphasize. I'd like to clarify this situation for you. #multitenancy In a system where everyone shares the same server, let's say one of your clients is running a Black Friday sale and sending thousands of requests to the system simultaneously, or trying to pull a very large and complex report. This situation causes the system our project is running on to lock its CPU at 100%. Ultimately, not only the noisy client on that server, but also all the other innocent clients whose applications are doing nothing at the time, will start to slow down. This officially creates a situation where everyone in the apartment building can't sleep if someone makes noise. #buildinpublic #dotnetpickaxe

if you haven't seen it yet, shrike labs just came out with a DC based power supply system for home / dev labs. Extremely cool stuff. youtube.com/watch?v=Ig7oZp…

@cb_doge xAI's approach to coding models is too chaotic. This works for free thinking but sucks for structured code.

ELON MUSK: "Grok is currently behind in coding. The reason I was late for this was that I was just in a giant sort of all hands on coding, going through all the things that need to happen to essentially exceed our competitors on coding, which I think we'll do. I feel we should probably get there by the middle of this year."

@nicksortor This is idiotic. We are making advances in neural networks all the time. This is a solution without a problem. The solution will be much better inference and models that use significantly less power.

🚨 JUST IN: President Trump has just announced an agreement with tech companies REQUIRING THEM to fully cover the cost of increased electricity production required for their data centers This has been a HUGE worry of Americans worried about their electric bills spiking as more and more data centers are built, so Trump stepped in. Trump says the new electricity generated from plants built by big tech firms will actually DRIVE DOWN energy prices for consumers, as excess power will go back into the grid. "This is an HISTORIC win for countless American families and will also make our electricity grids stronger and more resilient than ever before." "They will be getting a lot of excess electricity for the data centers where they build their own plants." "It's actually going to mean the prices come DOWN."

anyone else's isp getting salty about the amount of 600gb+ downloads and uploads we have to do for large models? i'm at 10tb for the month and climbing.

@MyLordBebo Amen.

EPSTEIN: This man’s message should be internalized by all I personally don’t care if it’s left or right or whatever

@BasedTorba That is not true. If you coded you would know that. AI is better than it used to be but will still create a giant mess with anything more than a class.

everyone should have their own local AI setup with multiple models for various things. Dense and thinking models and 400gb+ sizes are best. MOE models with huge training sets are great too. Lastly, don't forget to have vision and listening models.

@BrandonLuuMD My rhythm is 25 to 26 hours. If I just go to sleep when I want and wake up when I want, I cycle around the clock over a few weeks.

ADHD is closely linked to circadian rhythm dysfunction. Growing evidence suggests that targeting circadian misalignment can meaningfully improve symptoms. Grateful to Dr. Matt Walker for sharing our new study! frontiersin.org/journals/psych…

@elonmusk as a former video game programmer and current neural network architect/programmer, I'm not sure this is going to work the way most think it will. Creativity may be the last deeply human trait and the NN's can only mimic that.

MACROHARDER coming soon

@DataRepublican @SomeBitchIIKnow I programmed the game genie on the super ninetendo. Glad everyone liked it. :)

@SomeBitchIIKnow Oh I was so obsessed with inputting hex codes into the Game Genie and writing them all down

Yes we sure did.

@SQLServer I really love the new SSMS 22, extremely solid and intuitive. I hate the new connection dialog though, who thought that was a good idea? Can you at least make the old one optionally available or fix this one? #ssms

@VisualStudio this new visual studio 2026 is one of the best that has ever come out. A few bugs though that are driving me a bit crazy like installing or uninstalling an extension frequently leads to corruption of the extensions. 'ManagedProjectSystemPackage' and such did not load correctly.

🚀 Visual Studio 2026 is LIVE! Faster. Smarter. AI-native. ✅ 4,700 bugs fixed ✅ 300 new features ✅ Blazing-fast performance Learn more, upgrade now, and experience the future of development 👉 msft.it/6018tMUkM #VisualStudio2026 #AI

@tsoding I agree. Problem is everyone decided that all apps would be web (nobody still believes that) and just dropped the ball on native development. Winform was and is vastly superior to wpf and all the XAML variants. When windows had default look/feel we could do these things quickly.

You used to be able to just create a Native GUI App in 10 seconds. No Electron, no Game Engines, no Web frameworks. Just a lean fast .EXE produced in seconds. Works on any Windows machine WITHOUT Internet connection. Software Development is actually going backwards.

@MistralAI @arena can't wait to try it! my daily driver model for most things is glm 4.5 air (no quant full). Will be fun to compare.

Mistral Large 3 debuts as the #1 open source coding model on the @arena leaderboard. We'd love for you to try it! More on coding in a few days... 👀

@davepl1968 No it’s not.

It's true.

@davepl1968 winforms is still the best windows only tech. I wish MS gave it more love. XAML based solutions and wpf are insanely overly complex and try to solve a problem that doesn't exist (designers working alone without engineers).

I wrote a calculator in C# to learn WinForms, and rather than writing an expression evaluator, I just compiled the expression as code. It's a neat feature of .NET (live compilation of text to code)...

@nixcraft At least 128gb.

If someone were buying a MacBook Pro or a Thinkpad today for SWE/IT work, how much RAM would you recommend?

@davepl1968 and a windows 8 moment where they fired the people in charge of doing the horrible UI.

It's time for Microsoft to have another XPSP2 moment. No more AI, no more features. Just fixes. When I was working on Windows XP, Blaster hit. It was a big enough deal that we set aside all feature work. For the next several months, all we did was improve security. We didn't add "security features"; we fixed bugs. Lots of bugs. Until there weren't security bugs to fix anymore. Then we fixed the ones we didn't know about yet. Put more simply, we stopped trying to "add value" to the product through features that PMs thought users would like, and instead we focused on the things that had been important for a long time, but overlooked. Like performance and configurability today. Rather than trying to improve and add value to the system through new AI features -now-, I argue it's time for Microsoft to stabilize, improve, and make the system more performant. And more usable for power users. Just for one release. Just till it doesn't suck.