The DeepSeek reveal

If you have not heard about it yet, you will.

China claims the have developed an Artificial Intelligence (AI) model which is much less expensive to train.

The key here is “open-source” meaning that anyone can figure out how it works. Well, not anyone, but you know what I mean.

Still, a disturbing development if true for the makers of advanced hardware.

As mass-market GenAI has been a solution desperately seeking a problem since its announcement, this development is but another nail in the coffin.

1 Like

I think the big question is whether or not it is primarily trained with ChatGPT outputs. If it is trained from scratch, then that’s truly scary for US companies (china is doing more for much less cost). But if it builds on previous work, then further advances may still require hardware.

Even if it just copies part of the work, it’s still a very bad sign for companies like OpenAI… they spend the money and then someone else can copy it. Kind of ironic since AI is built on ingesting the work of a lot of human authors/artists without compensation.

2 Likes

You know what they say. “Garbage in=garbage out”.

3 Likes

Watching AI guys freak out about their meal ticket being in danger has been kinda funny.

1 Like

Just downloaded it on my phone last night and tried it out. Here’s an example of the kind of info it provided me when I asked for suggestions on MQTT topics:

While there is no universal standard for MQTT topics, there are best practices and common conventions that many developers and organizations follow to ensure consistency, scalability, and interoperability. Below, I’ll outline some widely accepted guidelines and examples of topic naming conventions.

### 1. General Best Practices
These are the most common recommendations for MQTT topic design:
1. Use a Clear Hierarchy:

    • Organize topics in a logical, tree-like structure.*
    • Example: domain/device-type/device-id/sensor-type*

2. Avoid Special Characters:

    • Use only alphanumeric characters, hyphens (-), underscores (_), and forward slashes (/).*
    • Avoid spaces, #, +, and other special characters unless used as wildcards.*

3. Be Consistent:

    • Use a consistent naming convention (e.g., lowercase, snake_case, or camelCase).*
    • Example: home/sensors/temperature/living-room*

4. Avoid Starting or Ending with a Slash:

    • Bad: /home/thermostat/*
    • Good: home/thermostat*

5. Use Wildcards Sparingly:

    • + (single-level wildcard): Matches one level in the hierarchy.*
    • # (multi-level wildcard): Matches all levels below.*

### 2. Common Conventions
While not formal standards, these conventions are widely used in the industry:

#### a. Domain-Based Structure
- Start with a high-level domain or application name.
- Example: home/sensors/temperature/living-room

#### b. Location-Based Structure
- Include location information in the topic hierarchy.
- Example: building/floor/room/device-type/device-id

#### c. Device-Centric Structure
- Focus on the device type and ID.
- Example: sensors/temperature/device-001

#### d. Action or Command Topics
- Use separate topics for commands and status updates.
- Example:

    • Command: home/actuators/light/living-room/command*
    • Status: home/actuators/light/living-room/status*

### 3. Industry-Specific Conventions
Some industries have developed their own conventions for MQTT topics:

#### a. Smart Home
- Example: home/{device-type}/{location}/{sensor-type}

    • home/sensors/temperature/living-room*
    • home/actuators/light/bedroom/switch*

#### b. Industrial IoT (IIoT)
- Example: factory/{machine-type}/{unit-id}/{sensor-type}

    • factory/motor/unit-1/status*
    • factory/sensors/temperature/zone-1*

#### c. Automotive
- Example: vehicle/{vehicle-id}/{sensor-type}

    • vehicle/car-001/speed*
    • vehicle/car-001/fuel-level*

#### d. Agriculture
- Example: farm/{field-id}/{sensor-type}

    • farm/field-1/moisture*
    • farm/field-1/temperature*

### 4. Example Topic Structures
Here are some examples of well-structured MQTT topics:

#### Smart Home
- home/sensors/temperature/living-room
- home/actuators/light/bedroom/switch
- home/security/door/front-door/status

#### Industrial IoT
- factory/motor/unit-1/status
- factory/sensors/temperature/zone-1
- factory/alarms/motor/unit-1/overheat

#### Automotive
- vehicle/car-001/speed
- vehicle/car-001/fuel-level
- vehicle/car-001/engine-temperature

#### Agriculture
- farm/field-1/moisture
- farm/field-1/temperature
- farm/field-1/irrigation/status

### 5. Tools and Resources
- Eclipse IoT: Provides guidelines for MQTT topic design in IoT systems.
- MQTT.org: Offers best practices and examples for MQTT topic naming.
- Industry-Specific Standards: Some industries (e.g., automotive, healthcare) may have their own guidelines for MQTT topics.

By following these conventions and best practices, you can create a scalable and maintainable MQTT topic structure for your application. Let me know if you need further assistance!

I have no complaints about it.

Dave

So I have been doing a bit of poking around. I think the most important thing to note about Deepseek is the terms of service. So running it locally you should not have issues though the terms of service are vague there. However if you are using their service from either the app store or the web you are subject to them. The big things are they data is store on servers in china both input and output. The data is subject to Chinese law. Also the terms are vague on who owns output from the model. I think as makers this one is a huge deal. Also It is unclear if someone uses Deepseek and comes up with a great business idea who owns that idea. I think this is a case of you have to know the cost of how your data is being used. I will be making a TikTok about this subject later today.

1 Like

Is there a formal group of AI enthusiasts that meetup at the shop on a regular basis? I am interested in collaborating with a few individuals on getting a use case for AI into service. I think it would be a cool collaborative between dprg, animatronics, electronics, computer science and even woodshop or welding.

If anyone else is interested please ping me.

Generative AI for Makers SIG Meeting
On the Event Calendar
https://calendar.dallasmakerspace.org/events/view/23926

1 Like

https://www.realclearpolitics.com/articles/2025/02/04/whos_afraid_of_jonathan_turley_chatgpt_for_one_152300.html

1 Like

The Chinese govt (or any govt for that matter) can already buy a lot of our data from data brokers, so I don’t feel like this makes that much of a difference.