The Download #011
February 7th, 2025
Dear Reader,
This week: Notes on DeepSeek, revisiting Make.com automations and more.
In the last few weeks, there’s been an explosion of activity and releases in the AI scene, due in part to DeepSeek R1’s disruptive impact in the tech, finance and political landscapes. While there remains a lot to digest in all those domains, I find myself resonating most with the following points:
Dario Amodei (Anthropic, and previous VP of Research at OpenAI), DeepSeek and Export Controls:
DeepSeek-V3 is not a unique breakthrough or something that fundamentally changes the economics of LLM’s; it’s an expected point on an ongoing cost reduction curve. What’s different this time is that the company that was first to demonstrate the expected cost reductions was Chinese.
Matthew Ingram (Tech Journalist), DeepSeek’s real magic doesn’t have anything to do with AI
But to me, the most interesting thing is that DeepSeek makes almost all of its work open source — anyone can download the model, and read the research papers that Wenfeng and his team have written about its development. HuggingFace, the French open-source repository for AI, said it already has 500 models available that were based on DeepSeek’s. In short, DeepSeek has done exactly what OpenAI said it was going to do when it was founded, but never actually did.
There are various versions and releases of DeepSeek to be found everywhere now, but I would still exercise caution about sending private and corporate information through it (as I would any LLM, to be honest), particularly as there are reports of the DeepSeek iOS app sending data unencrypted to ByteDance-controlled servers.
If you’re curious about what a ‘reasoning model’ is, compared to the usual ChatGPTs and Claude Sonnets, there’s an article further down to help you out.
Automations with Make.com
A while back I experimented with using Make.com (if you’re not familiar with it, it’s my pick for Tool of the Week, further down) with the idea of automating my social media posts. Doing this would hopefully take care of two issues at the time:
- I wanted to be able to “write once, post many”. In other words, take one post or idea, and then format it accordingly for the voice and audience of the different social networks where I was posting.
- I didn’t want to do this through a third-party social media automation posting tool like Hootsuite, or Buffer, where I’m charged for more than 3 social media accounts. Make already has the capability to post directly to them, so it should be simple. Right?
This is how far along I progressed with it:
16 modules of sweet, sweet automation.
You don’t need to be super-familiar with Make to understand this bit. Each icon represents a service, or module, that gets added to a scenario. Working from left-to-right and starting with Google Sheets, I’ll write a social media post which then gets refined it into something more readable and cohesive in the next module, by Perplexity.
When Perplexity finishes doing that, it hands off the result to a Router module, which you can then see branching out to other social network services.
From top-to-bottom:
- Facebook Pages gets a re-written post from Claude AI, accompanied by a Dall-E illustration.
- Instagram for Business gets an image, then a briefer post from ChatGPT with no link.
- Twitter gets a cut-down ChatGPT version of the post that stays within its character limit, with a link.
- LinkedIn gets a more professional version of the post, with bullet points and no emojis, also generated by Claude.
- Mastodon gets a similarly shortened version of the post, again via Claude, to support its 500 character limit, with a generated image from Dall-E to support it.
Since creating this automation the landscape of social media and how I choose to use it has dramatically shifted. These days, I’m hardly using Meta’s products (Facebook, Instagram, Threads) for anything personal, let alone professional. I’ve gone inactive on X, remaining only to make certain my namespace doesn’t get taken over, and to receive updates from local government and services that haven’t transitioned over to BlueSky or Mastodon.
So, that leaves only BlueSky, Mastodon and LinkedIn left, and you may notice that BlueSky is missing from my scenario above. That’s because it currently isn’t natively supported by Make.com, and a third-party plugin costs about $6 a month to enable it. No thanks - but it turns out Buffer, the solution I’d been trying to avoid using - does offer BlueSky…and if I can keep my Buffer account to only 3 social media accounts, I don’t have to sign up for a paid plan.
This has lead me to a more elegant solution, as I only technically need 3 accounts to post. Instead of having one giant automation scenario to rule them all, I’ve created smaller scenarios for similar platforms.
This one, for example, a) watches for anything new that’s been published on my website via RSS, b) uses Claude to rewrite it then c) pushes it to my Buffer queue.
RSS -> Claude - > Buffer
This next automation similarly takes the link and show notes from my podcast and turns them into a social media post with specific system and user instructions on what to do (and also what *not* to do). Provided I get my production done and uploaded in the morning, this runs automatically every day at noon to let people on my social networks that the newest episode is available.
Spotify (RSS) -> ChatGPT -> Buffer
So that’s where I’ve ended up. Smaller scenarios with custom instructions that can be easily replicated and tweaked for each social network. Buffer still plays a part, because it gives me an opportunity to check that what’s been posted is tonally on point.
For the moment, this works, and I’ll keep it in place for the rest of Q1. For more about Make, check out Tool of the Week below.
Latest Writing
Insight: The Evolution of AI in Marketing: From Spectacle to Strategy
By leveraging AI to handle data analysis, market research, and personalised content creation, marketing teams can focus on human creativity and strategic initiatives, leading to more effective and authentic campaigns.
Pages: Show Your Work: The Importance of AI Reasoning Models
AI reasoning models, a new generation of large language models designed to mimic human-like problem-solving, are transforming decision-making processes across industries by offering structured analysis and data-driven insights for complex problems.
Pages: AI and Tech Timeline Since DeepSeek R1
The release of DeepSeek’s R1 open-source AI model in January 2025 sent shockwaves through the tech industry, causing significant stock market volatility and raising questions about the future of AI development and competition between Chinese and American tech giants.
AI Tool of the Week: Make.com
I don’t think I need to try and convince you any further that Make.com is a tool worth checking out. There are templates available that can help you with digital marketing, general admin etc. - with more being added all the time.
Social media automation aside, just this week I was faced with a challenge in Google Workspace to set up an auto-reply system based on certain criteria coming into the main account inbox. I thought that Google’s AI assistant, Gemini, could help tackle the issue, only to be incorrect and disappointed. But I was able to create it in Make instead.
This automation auto-replies based on certain inbox criteria and manual tagging, creating a huge time savings.
If you’re interested in trying out Make.com automations for yourself, use the affiliate link below. If you’re curious about how you might be able to use this but not sure where to start, get in touch to book a session with me.
Inbox
Daisy the AI Granny
“Daisy is not a real grandmother but an AI bot created by computer scientists to combat fraud. Her task is simply to waste the time of the people who are trying to scam her. Using a mixture of ambivalence, confusion about how computers work and an eagerness to reminisce about her younger days, the “78 years young” Daisy draws sighs and snapping from fraudsters on the other end of the line.”
What a fantastic example of using AI to do the things that we shouldn’t have to.
If you’ve also got something you would like to share in the newsletter, get in touch!
Elsewhere
For daily, but sized tech and AI news updates, check out my new daily podcast “The Syntellicast”. It’s very nearly entirely generated using AI - more on how that all works in a future issue.
[
Feb 7th, 2025: OpenAI’s Euro…
Feb 7 · The Syntellicast
6:38
](https://open.spotify.com/show/621K1Phm2MUsIcdzBB6Trl)
And finally
Last Saturday we said farewell to Danté, our cat of nearly 19 years.
Dante made his mark online, appearing in i can haz cheezburger’s 2012 calendar, back when the internet was more about blogging and cat memes.
He was the ultimate lap cat and office buddy who has often sat alongside me writing this very newsletter, and frankly deserves a co-author credit. 🐾
See you next time.
Jim
Made with ❤️ in Valencia by Jim Christian. For feedback, please reach out to [email protected].
