The video effectively showcases a practical application of AI on consumer hardware, addressing a common user desire to run local models. It provides a clear demonstration of a user-friendly tool (LM Studio) and a specific model (Qwen3.5), making the complex topic of local AI accessible and actionable for a broader audience.
Summary
The creator demonstrates how to run local AI models on an older MacBook using LM Studio. They explore options like Ollama and then settle on LM Studio, highlighting its features and the Qwen3.5 model. The creator tests the model's capabilities with tasks like email generation, text summarization, PDF analysis, and code generation, noting its performance and limitations.
Structure
1Introduction of the old MacBook and the goal of running local AI models.
2Exploration of potential tools (Ollama, LM Studio) and selection of LM Studio.
Original caption
Running local models on old MacBook #aiforbeginners #localai #localllm
More from @easybits.tech.room
More breakout videos from this creator.
3
Demonstration of LM Studio's interface and the Qwen3.5 model.
4Testing the model with various tasks: email generation, text summarization, PDF analysis, and code generation.
5Discussion of the model's performance, limitations (factual errors, hallucinations), and the creator's surprise at its capabilities.
6Concluding remarks and a call to action to follow for future updates.
Product placement
Apple MacBook: acts - it is the device used to run local AI models. Removing it would change what happens in the video.
macOS: acts - it is the operating system on the MacBook. Removing it would change what happens in the video.
Ollama: acts - it is an inference engine for open models that the creator considered using. Removing it would change what happens in the video.
LM Studio: acts - it is a software application that allows users to run local AI models. The creator uses it to download and run models. Removing it would change what happens in the video.
Bionic: acts - it is an agent for open models released by LM Studio, designed for creativity, work, and code. It is presented as a feature of LM Studio. Removing it would change what happens in the video.
Qwen3.5-4B-GGUF: acts - it is a specific AI model that the creator downloads and uses within LM Studio. Removing it would change what happens in the video.
Qwen3.5: acts - it is an AI model with vision capabilities that the creator uses. Removing it would change what happens in the video.
Call to action
Follow me to not miss it
On-screen text
THIS IS MY OLD MACBOOK IT'S SIX YEARS OLD
IT HAS M1 CHIP AND 8 GIGABYTES OF UNIFIED MEMORY
AND I WANT TO TRY TO RUN SOME LOCAL MODELS ON IT
AND I'M QUITE BUSY TODAY SO
I DON'T WANT TO CONFIGURE
ANY COMPLEX INFERENCE ENGINE
I DON'T WANT TO WRITE ANY CODE
SO I WANT TO HAVE SOME OFF
THE SHELF
SOLUTION THAT I CAN JUST
INSTALL ON MY COMPUTER
DOWNLOAD THE MODEL AND JUST RUN IT
IF YOU DON'T KNOW ME MY NAME IS IVAN
I'M A COMPUTER SCIENCE TEACHER
FOUNDER OF ISABIT
AND I HELP ADULTS TO KEEP UP WITH TECH
LET'S RUN SOME MODELS
SO INITIALLY I WANTED TO GO WITH OLLAMA
THIS IS VERY POPULAR
INFERENCE ENGINE
BUT THEN I DECIDED THAT I WANT TO TRY SOMETHING ELSE
SO I WENT TO A LMSTUDIO
BECAUSE I USED THEM BEFORE
LM STUDIO
Go Bionic.
Bionic is LM Studio's agent for open models.
Natively local. Built for creativity, work, and code.
Copy download link
AGENTS POWER.
AT YOUR FINGERTIPS.
An agent for work and code.
Work with Bionic to create and edit documents. Every
change is automatically saved, so you can work with your
agent freely.
Bionic also excels at coding tasks, automations, and
computer control. Ask it for help with anything you need
done.
IT'S KIND OF AGENT WHERE
YOU CAN PLUG IN YOUR OWN
MODEL
IT HAS SOME BUILT IN TOOLS
SO YOU BASICALLY DON'T
NEED TO CONFIGURE ANYTHING
AND THEY HAVE VERY NICE
INTERFACE WITH A LIBRARY
WHERE YOU CAN LOOK FOR A
MODEL
YOU WANT TO DOWNLOAD AND
THEY ACTUALLY EVEN CHECK
IF YOU CAN RUN THE MODEL
ON YOUR HARDWARE
SO I WANTED TO GO WITH
QWEN
I HAVE 8 GIGABYTES OF
UNIFIED MEMORY
AND MY OPERATING SYSTEM
USES SOME
SO MOST LIKELY I CANNOT GO
ABOVE 4
5 GIGABYTES
SO I PROBABLY SHOULD GO WITH SOME SMALLER QWEN
MODEL
I DECIDED TO GO WITH QWEN3
3.5 4 BILLION PARAMETER
MODEL
AND I SEE THAT I CAN
DOWNLOAD EVEN DIFFERENT
QUANTIZED VERSIONS OF THIS
MODEL
I DECIDED TO GO WITH 4 BIT
QUANTIZATION
BECAUSE IT'S SMALLER AND
BECAUSE I'M NOT USING RAW LLM
AND I ACTUALLY PUT IT INTO
AN AGENT
I EXPECT THAT THIS SMALL
QUANTIZED VERSION
WILL STILL BE CAPABLE OF
DOING SOME STUFF
SO I DOWNLOADED THE MODEL
AND I WENT WITH SOME VERY
SIMPLE TASK
LIKE TEXT GENERATION
I ASKED IT TO WRITE AN EMAIL
TO MY BOSS THAT I'M SICK
AND NO SURPRISES IT DID
QUITE WELL
I ALSO ASKED IT TO
SUMMARIZE THE CHUNK OF
TEXT
WHICH IT ALSO HAVE DONE
QUITE EASILY
IT EVEN TRANSLATED ME
SOME TEXT FROM ENGLISH TO
RUSSIAN
AND EVEN THOUGH IT'S NOT
THE BEST TRANSLATION
IN THE WORLD IT STILL DIDN'T
MAKE ANY BIG MISTAKES
AND I DIDN'T HAVE ANY DOUBT
THAT IT WILL BE ABLE TO
GENERATE TEXT
I WANTED TO TRY IT ON
SOMETHING MORE SERIOUS
SO I UPLOADED PDF FILE AND
ASKED IT TO EXPLAIN ME
WHAT IS THIS?
IT WAS PROCESSING THE
TASK FOR MORE THAN TWO
MINUTES
IT ACTUALLY CONVERTED PDF
INTO PNG
AND THEN SENT IT TO A MODEL
AND BECAUSE QWEN 3.5 HAVE
VISION CAPABILITIES
IT WAS ABLE TO RECOGNIZE
WHAT THIS DOCUMENT IS
AND ACTUALLY CORRECTLY
EXTRACT INFORMATION
FROM THE DOCUMENT
I EVEN ASKED IT TO WRITE A
VERY SIMPLE
JAVASCRIPT FUNCTION AND
THAT HAVE DONE IT AS WELL
WHERE IT DIDN'T DO THIS
WELL IS FACTUAL
INFORMATION
I DISCONNECTED MY LAPTOP
FROM THE INTERNET
TO ELIMINATE ANY
WEB SEARCH TOOLS THAT
THIS AGENT MAY USE
WHO WAS THE PREVIOUS
PRESIDENT OF UNITED
STATES
AND IT SAID THAT BEFORE
DONALD TRUMP IT WAS
GEORGE H.W. BUSH
SO HE SKIPPED LIKE FIVE
OTHERS
BUT NO SURPRISE HERE
BECAUSE MODEL IS SMALL
AND QUANTIZED
SO OBVIOUSLY IT WILL
HALLUCINATE
OVERALL I WAS SURPRISED
THAT IT PERFORMED THAT
WELL
AND I ACTUALLY WANT TO
PUSH IT A LITTLE BIT HARDER
AND MAYBE EVEN RUN SOME
MULTI AGENTIC SYSTEM
AND I WILL SHARE RESULTS IN
THE NEXT EPISODES
SO FOLLOW ME TO NOT MISS IT
Transcript
This is my old MacBook it's six years old
And I want to try to run some local models on it
And I'm quite busy today so
I don't want to configure any complex inference engine
I don't want to write any code
So I want to have some off the shelf solution that I can just install on my computer
download the model and just run it
If you don't know me my name is Ivan
I'm a computer science teacher
Founder of easybits
And I help adults to keep up with tech
Let's run some models
So initially I wanted to go with Ollama
This is very popular inference engine
But then I decided that I want to try something else
So I went to a LM Studio because I used them before
LM Studio
Go Bionic.
Bionic is LM Studio's agent for open models.
Natively local. Built for creativity, work, and code.
Copy download link
Agentic power. At your fingertips.
An agent for work and code.
Work with Bionic to create and edit documents. Every change is automatically saved, so you can work with your agent freely.
Bionic also excels at coding tasks, automations, and computer control. Ask it for help with anything you need done.
It's kind of agent where you can plug in your own model
It has some built in tools
So you basically don't need to configure anything
And they have very nice interface with a library
Where you can look for a model
You want to download and they actually even check
if you can run the model on your hardware
So I wanted to go with Qwen
I have 8 gigabytes of unified memory
And my operating system uses some
So most likely I cannot go above 4
5 gigabytes
So I probably should go with some smaller Qwen model
I decided to go with Qwen3
3.5 4 billion parameter model
And I see that I can download even different quantized versions of this model
I decided to go with 4 bit quantization
Because it's smaller and because I'm not using raw LLM
And I actually put it into an agent
I expect that this small quantized version
will still be capable of doing some stuff
So I downloaded the model
And I went with some very simple task
like text generation
I asked it to write an email to my boss that I'm sick
And no surprises it did quite well
I also asked it to summarize the chunk of text
Which it also have done quite easily
It even translated me some text from English to Russian
And even though it's not the best translation in the world it still didn't make any big mistakes
And I didn't have any doubt that it will be able to generate text
I wanted to try it on something more serious
So I uploaded PDF file and asked it to explain me
What is this?
It was processing the task for more than two minutes
It actually converted PDF into PNG
And then sent it to a model
And because Qwen 3.5 have vision capabilities
It was able to recognize what this document is
And actually correctly extract information from the document
I even asked it to write a very simple JavaScript function and that have done it as well
Where it didn't do this well is factual information
I disconnected my laptop from the internet
To eliminate any web search tools that this agent may use
Who was the previous president of United States
And it said that before Donald Trump it was George H.W. Bush
So he skipped like five others
But no surprise here
Because model is small and quantized
So obviously it will hallucinate
Overall I was surprised that it performed that well
And I actually want to push it a little bit harder
And maybe even run some multi agentic system
And I will share results in the next episodes
So follow me to not miss it