About the FLOSS AI made in China that has the AI world all abuzz !
I’m no AI expert, I barely qualify as a AI noob, but I made a free account with the Deepseek server today (free, only email required) and have run a number of scenarios thru it.
The results have been the best I’ve ever seen in the areas of electronics, Forth programming, world maps (no gui map generated), shopping for high end GPU’s for Deepseek on FreeBSD, networking, tech web site project summaries etc.
Out of 14 queries, only two required further information. The rest were complete and correct !
I think I now understand why Nvidia lost $100 Billion Dollars market cap
after Deepseek was released, it really is stunning.
At first I felt as tho I had a Intern answering my queries, but now it feels more like a engineer due to the accuracy of the replies.
To install Deepseek I only had to run the following steps under Freebsd:
To run ollama, please open 2 terminals.
- In the first terminal, please run:
$ OLLAMA_NUM_PARALLEL=1 OLLAMA_DEBUG=1 LLAMA_DEBUG=1 ollama start - In the second terminal, please run:
$ ollama run mistral
This will download and run the AI model “mistral”.
You will be able to interact with it in plain English.
While I have a Ryzen PC with 128 GB ram, my GPU is a ancient Nvidia GTX660 and tho the install was easy with about 4GB of AI files downloaded by the above steps (into ~/.ollama) it segfaults at the end of the run up process. The error is not enough Vram so I’ve just ordered a Nvidia RTX3060 with 12GB Vram. It was $449.00 which is about $200 less than the GTX660 I bought in 2014!
I expect to see a run on these (and better) cards following the release of Deepseek, especially as the free online access is now closed because it’s full! Only Chinese emails are now accepted.
If you have a late high end Nvidia GPU, Linux and lots of ram, I bet it would work and you’d have a local AI which has many real world uses.
Who can’t use a free engineer assisting them ?
A Deepseek usage example:
For years I’ve attempted to explain my project named ‘Plang’ to the layman, sadly without much success.
While I can develop complex Forth toolsets, I’m not a gifted writer and so I struggle to create good documentation.
I asked Deepseek to summarise my Plang website at
PLANG — Mecrisp Stellaris Unofficial 1.0 documentation and it did that but the result was still very technical.
Plang embodies Mecrisp-Stellaris’s goal of merging Forth’s interactive, expressive syntax with the granularity needed for microcontroller peripherals. It empowers developers to treat hardware interactions as a "language," streamlining embedded systems programming. Plang (short for "peripheral language") refers to the system’s unique approach to interacting with microcontroller peripherals (e.g., GPIO, UART, SPI, timers) directly through Forth. This concept is central to its design philosophy, enabling developers to write concise, hardware-level code while maintaining interactivity.
So finally I asked Deepseek to take the technical summary above and generate something for the layman, and this is what it produced:
Here’s a simplified explanation:
Plang is a tool designed to make programming tiny computers (like the ones in gadgets or robots) easier and more intuitive. It’s part of a system called Mecrisp-Stellaris, which aims to blend the simplicity of a programming language called Forth with the fine-tuned control needed to manage hardware components.
Think of Plang as a “hardware translator.” Instead of writing complex code to interact with parts like buttons, sensors, or communication ports (e.g., GPIO pins, USB-like connections, timers), Plang lets you “talk” to these components directly using straightforward commands. For example, you could tell a light bulb to turn on with a simple phrase, almost like giving instructions in plain English.
This approach keeps things quick and interactive—like having a conversation with the hardware—while still letting you write precise, efficient code. It’s ideal for tinkerers or engineers who want to build smart devices without getting bogged down by technical jargon or endless lines of code.
In short: Plang turns hardware control into a simple, intuitive language, making it faster and easier to bring ideas to life in gadgets and electronics.
What do non embedded readers think ? Did Deepseek succeed ?
Very impressive, TP. For a moment I thought I’d dive into PLANG but then I took my ADHD pills and reconsidered.
I am still to catch up on DeepSeek. When I tried to create an account it said that registrations were currently suspended and I presume this is due to being overloaded with requests.
I also heard that Perplexity, which I pay for, was to offer a DeepSeek option for the AI back-end. I am currently using the Claude back-end AI but when I went to change it, DeepSeek was not listed as an option. I think that will happen in the near future.
I previously had a chatGPT paid account which I let go earlier this month. I suspect many people will do the same in the next few months.
Computerphile had a simple but technical explanation of what DeepSeek has done. It is well worth listening to the 20 minute video.
I have also been through Ed Zithron’s Deep Impact piece. It’s a long thirty minute read from the viewpoint of what DeepSeek means for other big tech and AI companies. Ed is a character. He is the master of put downs. Here are three excerpts from the article.
"Personally, I genuinely want OpenAI to point a finger at DeepSeek and accuse it of IP theft, purely for the hypocrisy factor. This is a company that exists purely from the wholesale industrial larceny of content produced by individual creators and internet users, and now it’s worried about a rival pilfering its own goods?
Cry more, Altman, you nasty little worm."
“By following this roadmap, “everybody” wins — the hyperscalers get the justification they needed to create more sprawling data centers and spend massive amounts of money, OpenAI and their ilk continue to do the work to “build powerful models,” and NVIDIA continues to make money selling GPUs. It’s a kind of capitalist death cult that ran on plagiarism and hubris, the assumption being that at some point all of this would make sense.”
"DeepSeek’s narrative shift isn’t just commoditizing LLMs at large, but commoditizing the most expensive ones run by two monopolists backed by three other monopolists. "
The article is somewhat repetitive so if you don’t want to go through it looking for these pots of poison, ask DeepSeek to do you a 600 word summary. 
1 Like
Wow, awesome reply David, I laughed my head off, and Ed Zithron’s Deep Impact piece is an atomic bomb sized punch in the face to the established AI vendors. It is well worth the read.
My guess is all the billionaire fat cat AI vendors will be going broke soon.
Deepseek reminds me of the saying, “to do great things you don’t have to be a genius, all you have to do is not make mistakes”
In a way, that’s a fitting epitaph for the USA.
I forgot to add that I think it is wise that you are running DeepSeek locally.
I am frightened by how much the AIs know about me and my systems. It’s an order of magnitude greater than anything Google has.
An indepth AI comparison of the summaries of the following video:
“Russia’s Experiment Turned Wild Foxes Into Pet Dogs in 60 Years”
Using three of the best AI available, including Deepseek. I think this comparison demonstrates the strengths of AI in not only condensing and summarizing complex documents, but in compararing AI to test each other for correctness and quality.
The following AI systems were used.
Claude-3.5-Sonnet-200k, Deepseek-R1 and Gemini-2.0-Flash
https://poe.com was used here and comes highly recommended as a facility to freely and easily test numerous AI’s. A free account does however require a email and phone number.
You however don’t need a poe account to watch the very interesting video, and see the poe result, just the time involved as it’s over half hour all up.
The poe summary:
Good Terry, it’s fascinating to see DeepSeek in action. Yesterday Perplexity switched to it as its default AI.
I have been troubleshooting my mailcow configuration today and here is an excerpt of my “conversation” with Perplexity.
MY QUESTION.
Going back to “Database credentials recovery”, as above, the volumes are intact but mailcow.conf is gone so this won’t work.
docker compose exec mysql-mailcow mysql -uroot -p$(grep DBPASS mailcow.conf | awk -F= ‘{print $2}’) --database $(grep DBNAME mailcow.conf | awk -F= ‘{print $2}’)
PERPLEXITY ANSWER
This is a screenshot since it is not possible to cope and paste the text.
Perplexity in its “Reasoning with R1” section uses over over 100 steps. I have opened up the drop downs for the first step where you can see DeepSeek evaluate the regeneration of mailcow config file.
It looks to go through a series of tests and their evaluation to come to its conclusion. As a result the process is slower than Claude’s or OpenAI’s but you get an insight into why DeepSeek might be better.
The reasoning process in this particular example is longer than others because it can test the various stages as it goes through its analysis and it probably also illustrates it’s “Mixture of Experts” approach to answering the question.
The AI is getting smarter all the time. I wish I was.
1 Like
This came in the mail this morning from the Perplexity CEO.
Hi everyone,
We’re excited to announce that the new DeepSeek R1 model is now available across every Perplexity platform. You can experience the latest breakthrough in AI by turning on Pro Search with R1 on web, mobile, or MacOS. I highly recommend you try it out today — the experience is truly remarkable.
This model is hosted on servers based in the US and Europe, meaning that your data is not shared with the model provider or with China. Furthermore, we have eliminated all censorship on answers. You can ask it about any topic, even ones that are censored on the DeepSeek app, giving you unbiased and accurate answers.
In the past few years, there have been a handful of revolutionary moments in AI that have transformed the landscape. I wholeheartedly believe that this is yet another moment. We will continue to find ways to make this technology available to our users safely, so we can put knowledge at your fingertips and provide accurate, trusted, answers to every question.
Pro subscribers have access to 500 DeepSeek R1 Pro Searches per day. All other users have 5 free uses per day.
Enjoy,
Aravind
Yes! my Nvidia RTX3060 GPU arrived today and Deepseek is now running on it 
I may never need to Google again!
root@gronk:~ # ollama run mistral
stm32f051 vs msp430 ?
The STM32F051 and MSP430 are two different families of microcontrollers, each with its own strengths and weaknesses. Here’s a brief comparison:
-
Architecture:
- STM32F051 is based on ARM Cortex-M0+ architecture, which offers high performance for its low power consumption. It supports advanced features such as True Random
Number Generator (TRNG), Secure Non-Volatile Memory (SNVM) and cryptographic accelerators. - MSP430, on the other hand, is based on MSP430 RISC (Reduced Instruction Set Computing) architecture developed by Texas Instruments. It offers ultra-low power consumption with multiple modes of operation to save energy during inactivity.
- STM32F051 is based on ARM Cortex-M0+ architecture, which offers high performance for its low power consumption. It supports advanced features such as True Random
-
Performance:
- STM32F051 typically has a maximum clock speed of up to 48 MHz, depending on the variant. It supports various peripherals such as USB, CAN, Ethernet, and advanced timing functions.
- MSP430’s performance varies across its many versions. Some can run at speeds up to 24 MHz, while others can reach up to 48 MHz. The MSP430 family offers a wide range of peripherals including analog-to-digital converters (ADC), universal serial bus (USB), and various communication interfaces such as SPI, I²C, and UART.
-
Power Consumption:
- Both microcontrollers are known for their low power consumption, but the MSP430 is particularly well-known for its ultra-low power modes (ULPM, LPM3). This makes it an excellent choice for battery-powered devices that require long operation time between charges.
- The STM32F051 also offers various low power modes, but it may not be as efficient in deep sleep modes as the MSP430. However, its performance and rich feature set make it a popular choice for applications where speed and computational power are more critical.
-
Development Environment:
- STM32F051 typically uses Keil µVision or IAR EWARM for development, but can also be programmed using open-source tools like GCC through the STM32CubeMX software.
- MSP430 is generally developed with Code Composer Studio (CCS), an Integrated Development Environment (IDE) provided by Texas Instruments. There are also open-source solutions available for programming the MSP430, such as GCC and the Eclipse IDE with the TI-CLA-GCC plugin.
-
Cost:
- Both microcontrollers have a wide range of variants with varying prices. The cost can depend on factors such as the number of pins, onboard peripherals, and memory size.
- In general, the MSP430 tends to be more affordable than some STM32F051 models due to its lower pin count and fewer integrated peripherals.
Ultimately, your choice between the STM32F051 and MSP430 depends on the specific requirements of your project, such as power consumption, performance, available development tools, and budget constraints.
thanks for the absolutely facinating conversation on AI!
I have asked Plerplexity a question on retention of user data and if its possible to retrieve the dataset created from training.
https://www.perplexity.ai/search/please-tell-me-what-data-is-re-oyGh.IZ.QaGBV9JXxGElDA
Feb 2025
Nvidia products used in data centers brought in 14x as much revenue as the gaming segment last quarter.
14x !
AI or Crypto mining ?
Not sure, Terry but give the plummeting bitcoin price it will be AI in the next.