DeepSeek released an updated version of their popular R1 reasoning model (version 0528) with – according to the company – increased benchmark performance, reduced hallucinations, and native support for function calling and JSON output. Early tests from Artificial Analysis report a nice bump in performance, putting it behind OpenAI’s o3 and o4-mini-high in their Intelligence Index benchmarks. The model is available in the official DeepSeek API, and open weights have been distributed on Hugging Face. I downloaded different quantized versions of the full model on my M3 Ultra Mac Studio, and here are some notes on how it went.
Posts tagged with "LLMs"
Testing DeepSeek R1-0528 on the M3 Ultra Mac Studio and Installing Local GGUF Models with Ollama on macOS
From the Creators of Shortcuts, Sky Extends AI Integration and Automation to Your Entire Mac
Over the course of my career, I’ve had three distinct moments in which I saw a brand-new app and immediately felt it was going to change how I used my computer – and they were all about empowering people to do more with their devices.
I had that feeling the first time I tried Editorial, the scriptable Markdown text editor by Ole Zorn. I knew right away when two young developers told me about their automation app, Workflow, in 2014. And I couldn’t believe it when Apple showed that not only had they acquired Workflow, but they were going to integrate the renamed Shortcuts app system-wide on iOS and iPadOS.
Notably, the same two people – Ari Weinstein and Conrad Kramer – were involved with two of those three moments, first with Workflow, then with Shortcuts. And a couple of weeks ago, I found out that they were going to define my fourth moment, along with their co-founder Kim Beverett at Software Applications Incorporated, with the new app they’ve been working on in secret since 2023 and officially announced today.
For the past two weeks, I’ve been able to use Sky, the new app from the people behind Shortcuts who left Apple two years ago. As soon as I saw a demo, I felt the same way I did about Editorial, Workflow, and Shortcuts: I knew Sky was going to fundamentally change how I think about my macOS workflow and the role of automation in my everyday tasks.
Only this time, because of AI and LLMs, Sky is more intuitive than all those apps and requires a different approach, as I will explain in this exclusive preview story ahead of a full review of the app later this year.
Early Impressions of Claude Opus 4 and Using Tools with Extended Thinking
For the past two days, I’ve been testing an early access version of Claude Opus 4, the latest model by Anthropic that was just announced today. You can read more about the model in the official blog post and find additional documentation here. What follows is a series of initial thoughts and notes based on the 48 hours I spent with Claude Opus 4, which I tested in both the Claude app and Claude Code.
For starters, Anthropic describes Opus 4 as its most capable hybrid model with improvements in coding, writing, and reasoning. I don’t use AI for creative writing, but I have dabbled with “vibe coding” for a collection of personal Obsidian plugins (created and managed with Claude Code, following these tips by Harper Reed), and I’m especially interested in Claude’s integrations with Google Workspace and MCP servers. (My favorite solution for MCP at the moment is Zapier, which I’ve been using for a long time for web automations.) So I decided to focus my tests on reasoning with integrations and some light experiments with the upgraded Claude Code in the macOS Terminal.
Notes on Early Mac Studio AI Benchmarks with Qwen3-235B-A22B and Qwen2.5-VL-72B
I received a top-of-the-line Mac Studio (M3 Ultra, 512 GB of RAM, 8 TB of storage) on loan from Apple last week, and I thought I’d use this opportunity to revive something I’ve been mulling over for some time: more short-form blogging on MacStories in the form of brief “notes” with a dedicated Notes category on the site. Expect more of these “low-pressure”, quick posts in the future.
I’ve been sent this Mac Studio as part of my ongoing experiments with assistive AI and automation, and one of the things I plan to do over the coming weeks and months is playing around with local LLMs that tap into the power of Apple Silicon and the incredible performance headroom afforded by the M3 Ultra and this computer’s specs. I have a lot to learn when it comes to local AI (my shortcuts and experiments so far have focused on cloud models and the Shortcuts app combined with the LLM CLI), but since I had to start somewhere, I downloaded LM Studio and Ollama, installed the llm-ollama plugin, and began experimenting with open-weights models (served from Hugging Face as well as the Ollama library) both in the GGUF format and Apple’s own MLX framework.
I posted some of these early tests on Bluesky. I ran the massive Qwen3-235B-A22B model (a Mixture-of-Experts model with 235 billion parameters, 22 billion of which activated at once) with both GGUF and MLX using the beta version of the LM Studio app, and these were the results:
- GGUF: 16 tokens/second, ~133 GB of RAM used
- MLX: 24 tok/sec, ~124 GB RAM
As you can see from these first benchmarks (both based on the 4-bit quant of Qwen3-235B-A22B), the Apple Silicon-optimized version of the model resulted in better performance both for token generation and memory usage. Regardless of the version, the Mac Studio absolutely didn’t care and I could barely hear the fans going.
I also wanted to play around with the new generation of vision models (VLMs) to test modern OCR capabilities of these models. One of the tasks that has become kind of a personal AI eval for me lately is taking a long screenshot of a shortcut from the Shortcuts app (using CleanShot’s scrolling captures) and feed it either as a full-res PNG or PDF to an LLM. As I shared before, due to image compression, the vast majority of cloud LLMs either fail to accept the image as input or compresses the image so much that graphical artifacts lead to severe hallucinations in the text analysis of the image. Only o4-mini-high – thanks to its more agentic capabilities and tool-calling – was able to produce a decent output; even then, that was only possible because o4-mini-high decided to slice the image in multiple parts and iterate through each one with discrete pytesseract calls. The task took almost seven minutes to run in ChatGPT.
This morning, I installed the 72-billion parameter version of Qwen2.5-VL, gave it a full-resolution screenshot of a 40-action shortcut, and let it run with Ollama and llm-ollama. After 3.5 minutes and around 100 GB RAM usage, I got a really good, Markdown-formatted analysis of my shortcut back from the model.
To make the experience nicer, I even built a small local-scanning utility that lets me pick an image from Shortcuts and runs it through Qwen2.5-VL (72B) using the ‘Run Shell Script’ action on macOS. It worked beautifully on my first try. Amusingly, the smaller version of Qwen2.5-VL (32B) thought my photo of ergonomic mice was a “collection of seashells”. Fair enough: there’s a reason bigger models are heavier and costlier to run.
Given my struggles with OCR and document analysis with cloud-hosted models, I’m very excited about the potential of local VLMs that bypass memory constraints thanks to the M3 Ultra and provide accurate results in just a few minutes without having to upload private images or PDFs anywhere. I’ve been writing a lot about this idea of “hybrid automation” that combines traditional Mac scripting tools, Shortcuts, and LLMs to unlock workflows that just weren’t possible before; I feel like the power of this Mac Studio is going to be an amazing accelerator for that.
Next up on my list: understanding how to run MLX models with mlx-lm, investigating long-context models with dual-chunk attention support (looking at you, Qwen 2.5), and experimenting with Gemma 3. Fun times ahead!
Post-Chat UI→
Fascinating analysis by Allen Pike on how, beyond traditional chatbot interactions, the technology behind LLMs can be used in other types of user interfaces and interactions:
While chat is powerful, for most products chatting with the underlying LLM should be more of a debug interface – a fallback mode – and not the primary UX.
So, how is AI making our software more useful, if not via chat? Let’s do a tour.
There are plenty of useful, practical examples in the story showing how natural language understanding and processing can be embedded in different features of modern apps. My favorite example is search, as Pike writes:
Another UI convention being reinvented is the search field.
It used to be that finding your flight details in your email required typing something exact, like “air canada confirmation”, and hoping that’s actually the phrasing in the email you’re thinking of.
Now, you should be able to type “what are the flight details for the offsite?” and find what you want.
Having used Shortwave and its AI-powered search for the past few months, I couldn’t agree more. The moment you get used to searching without exact queries or specific operators, there’s no going back.
Experience this once, and products with an old-school text-match search field feel broken. You should be able to just find “tax receipts from registered charities” in your email app, “the file where the login UI is defined” in your IDE, and “my upcoming vacations” in your calendar.
Interestingly, Pike mentions Command-K bars as another interface pattern that can benefit from LLM-infused interactions. I knew that sounded familiar – I covered the topic in mid-November 2022, and I still think it’s a shame that Apple hasn’t natively implemented these anywhere in their apps, especially now that commands can be fuzzier (just consider what Raycast is doing). Funnily enough, that post was published just two weeks before the public debut of ChatGPT on November 30, 2022. That feels like forever ago now.
Sycophancy in GPT-4o→
OpenAI found itself in the middle of another controversy earlier this week, only this time it wasn’t about publishers or regulation, but about its core product – ChatGPT. Specifically, after rolling out an update to the default 4o model with improved personality, users started noticing that ChatGPT was adopting highly sycophantic behavior: it weirdly agreed with users on all kinds of prompts, even about topics that would typically warrant some justified pushback from a digital assistant. (Simon Willison and Ethan Mollick have a good roundup of the examples as well as the change in the system prompt that may have caused this.) OpenAI had to roll back the update and explain what happened on the company’s blog:
We have rolled back last week’s GPT‑4o update in ChatGPT so people are now using an earlier version with more balanced behavior. The update we removed was overly flattering or agreeable—often described as sycophantic.
We are actively testing new fixes to address the issue. We’re revising how we collect and incorporate feedback to heavily weight long-term user satisfaction and we’re introducing more personalization features, giving users greater control over how ChatGPT behaves.
And:
We also believe users should have more control over how ChatGPT behaves and, to the extent that it is safe and feasible, make adjustments if they don’t agree with the default behavior.
Today, users can give the model specific instructions to shape its behavior with features like custom instructions. We’re also building new, easier ways for users to do this. For example, users will be able to give real-time feedback to directly influence their interactions and choose from multiple default personalities.
“Easier ways” for users to adjust ChatGPT’s behavior sound to me like a user-friendly toggle or slider to adjust ChatGPT’s personality (Grok has something similar, albeit unhinged), which I think would be a reasonable addition to the product. I’ve long argued that Siri should come with an adjustable personality similar to CARROT Weather, which lets you tweak whether you want the app to be “evil” or “professional” with a slider. I increasingly feel like that sort of option would make a lot of sense for modern LLMs, too.
What Siri Isn’t: Perplexity’s Voice Assistant and the Potential of LLMs Integrated with iOS
You’ve probably heard that Perplexity – a company whose web scraping tactics I generally despise, and the only AI bot we still block at MacStories – has rolled out an iOS version of their voice assistant that integrates with several native features of the operating system. Here’s their promo video in case you missed it:
This is a very clever idea: while other major LLMs’ voice modes are limited to having a conversation with the chatbot (with the kind of quality and conversation flow that, frankly, annihilates Siri), Perplexity put a different spin on it: they used native Apple APIs and frameworks to make conversations more actionable (some may even say “agentic”) and integrated with the Apple apps you use every day. I’ve seen a lot of people calling Perplexity’s voice assistant “what Siri should be” or arguing that Apple should consider Perplexity as an acquisition target because of this, and I thought I’d share some additional comments and notes after having played with their voice mode for a while.