Deepseek: Why Trust Any Smart Software?
October 16, 2025
This essay is the work of a dumb dinobaby. No smart software required.
We have completed our work on my new book “The Telegram Labyrinth.” In the course of researching and writing about Pavel Durov’s online messaging system, we learned one thing: Software is not what it seems to the user. Most Telegram users believe that Telegram is end to end encrypted. It is, but only if the user goes through some hoops. The vast majority of users don’t go through hoops. Those millions upon millions of users know much about the third-party bots chugging away in Groups and Channels (public and private). Even fewer users realize that a service charge is applied to each monetary transaction in the Telegram system. That money flows to the GOAT (greatest of all time) technical wizard, Pavel Durov and some close associates. Who knew?
I read “The Demonization of Deepseek: How NIST Turned Open Science into a Security Scare.” The write up focuses on a study or analysis conducted by what used to be the National Bureau of Standards. (I loved those traffic jams on Quince Orchard Road in Gaithersburg, Maryland.) The software put under the NIST (National Institute of Science & Technology) is the China-linked Deepseek smart software.
The cited article discusses the NIST study. Let’s see what it says about the China-linked artificial intelligence system. Presumably Deepseek did more with less; that is, the idea was to demonstrate that Chinese innovation could make US methods of large language models. The result would be better, faster, and cheaper. Cheap has a tendency to win in some product and service categories. Also, “good enough” is a winner in today’s market. (How about the reliability of some of those 2025 automobiles and trucks?)
The write up says:
NIST’s recent report on Deepseek is not a neutral technical evaluation. It is a political hit piece disguised as science. There is no evidence of backdoors, spyware, or data exfiltration. What is really happening is the U.S. government using fear and misinformation to sabotage open science, open research, and open source. They are attacking gifts to humanity with politics and lies to protect corporate power and preserve control. Deepseek’s work is a genuine contribution to human knowledge, and it is being discredited for reasons that have nothing to do with security.
Okay, that’s clear.
Let’s look at how the cited write up positions Deepseek:
Deepseek built competitive AI models. Not perfect, but impressive given their budget. They spent far less than OpenAI or Anthropic and still achieved near-frontier performance. Then they open-sourced everything under Apache 2.0.
The point of the write up is that analysis has been politicized. This is an interesting allegation. I am not confident that any “objective” analysis is indeed without spin. Remember those reports about smoking cigarettes and the work of the Tobacco Institute. (I am a dinobaby, but I remember.)
The write up does identify three concerns a user of Deepseek should have. Let me quote from the cited article:
- Using Deepseek’s API: If you send sensitive data to Deepseek’s hosted service, that data goes through Chinese infrastructure. This is a real data sovereignty issue, the same as using any foreign cloud provider.
- Jailbreak susceptibility: If you’re building production applications, you need to test ANY model for vulnerabilities and implement application-level safeguards. Don’t rely solely on model guardrails. Also – use an inference time guard model (such as LlamaGuard or Qwen3Guard) to classify and filter both prompts and responses.
- Bias and censorship: All models reflect their training data. Be aware of this regardless of which model you use.
Let me offer several observations:
- Most people are unaware of what can be accomplished from software use. Assumptions about what it does and does not do are dangerous. We have tested Deepseek running locally. It is okay. This means it can do some things well like translate a passage in English into German. It has no clue about timely issues because most LLMs are not updated in near real time. Some are, but others are not. Who needs timely information when cheating on a high school essay? Answer: no one.
- The write up focuses on Deepseek, but its implications are much more broad. I think that the mindless write ups from consulting firms and online magazines is a very big problem. Critical thinking is just not the common. It is a problem in the US but other countries have this blind spot as well.
- The idea that political perceptions alter what should be an objective analysis is troubling to me. I have written a number of reports for government agencies; for example, a report about Japan’s obsession with a database industry for the Office of Technology Assessment. Yep, I am a dinobaby remember. I may have been right or wrong in my report, but I was not influenced by any political concept or actor. I could have been because I did a stint in the office of Admiral / Congressman Craig Hosmer. My OTA work was not part of the “game” for me.
Net net: Trust is important. I think it is being eroded. I also believe that there are few people who present information without fear or favor. Now here’s the key part of my perception: One cannot trust smart software or any of the programmer assembled, hidden threshold, and masked training methods that go into these confections. More critical thinking is needed. A deceptive business practice if well crafted cannot be perceived. Remember Telegram Messenger is 13 years young and users of the system don’t have much awareness of bots, mini apps, and dapps. What don’t people know about smart software?
Stephen E Arnold, October 16, 2025
Comments
Got something to say?
-
- Subscribe to Beyond Search
Feature archive
News archive
-
