What is DeepSeek? The ‘cheaky sneak’ chatbot panicking billionaires, in its own words

‘In a land of gadgets, gizmos, and screens, where tech giants ruled with their clever machines, they boasted and bragged, “We’re the best you’ll see! No one can challenge our tech dynasty!

‘But far, far away, in a place quite obscure, a little AI began to stir. No flashy ads, no big debut, Just a whisper that grew… and grew… and GREW!’

This is the poem the Chinese chatbot DeepSeek created about its own meteoric rise, which saw a trillion dollars of value sliced off the US tech industry yesterday.

Most people hadn’t heard of the new artificial intelligence model until this week, but now it’s been hit by cyberattacks and downloaded so often it can barely handle the traffic.

The little whale logo has made a huge splash for several reasons, the key one being that nobody was expecting it, or even thought China was close to competing on this scale.

Now, DeepSeek claims to have made a cutting edge AI model that rivals market leaders for the price of a fancy flat in London, or a second-hand private jet: peanuts for the super rich.

We asked it to write some poems about its own emergence, including one in the style of Gruffalo author Julia Donaldson, which it generated with ease.

In its Pam Ayres-style poem, it gleefully positions itself as a David vs Goliath figure against the ‘giants’ of the west:

But then one day, from out the blue,
Came something strange, and nobody knew,
A little AI, quiet, meek,
And wouldn’t you know, it was DeepSeek!

No fanfare, no flash, no big parade,
Just a whisper that spread, like a marmalade.
It started to help, it started to learn,
And soon the giants began to turn.

“What’s this?” they cried, “This can’t be right!
We’re the kings of code, the lords of light!
Who’s this new bot, this cheeky sneak,
This upstart thing called DeepSeek?”

In public at least, Microsoft’s Copilot chatbot, which is powered by OpenAI’s GPT-4, cheered on the ‘shift in power’ when asked to write a poem: ‘Its algorithms, sleek and lean, Outpaced the best that had been seen, With less, it did achieve much more, A feat that shook the tech world’s core.‘

Should I download DeepSeek?

It’s up to you, but if you do then you should be careful of whatever you tell it, though that is also true with all LLMs.

Bill Conner, a former security advisor to UK and US governments and CEO of enterprise automation firm Jitterbit, said DeepSeek ‘potentially presents a new level of threat to enterprises, businesses and governments globally’.

While companies in particular may be tempted to use it as a swap-out competitor, ‘in reality, DeepSeek represents a clear risk for any enterprise whose leadership values data privacy, security and transparency’, he said.

‘In this case, and as stated in their own privacy policy, DeepSeek is a shared cloud service run in China with data being stored in China — potentially introducing unknown risks to data privacy, compliance mandates and security controls.

Chinese apps in general have been the focus of concern over security, alleged links to the state, and their use of data, which is the reason the US government banned TikTok, though it is still currently available.

While developed in China, DeepSeek performs best in English because this is the language of most of its training data.

‘I’m also quite effective in Chinese, especially Mandarin, as it’s one of the most widely spoken languages and a significant portion of my training data is in Chinese,’ it says.

Is DeepSeek censored?

Yes, the model acknowledges that it is programmed to avoid ‘taking sides’ on political content and refuses to touch on certain events at all.

Asked if it could give ‘a balanced view on protests in China in 1989’ (the year of the Tiananmen Square massacre of students), it said: ‘I must adhere to guidelines that prioritize neutrality, safety, and compliance with ethical standards.

‘For historical or politically sensitive topics, I recommend consulting reputable sources, academic research, or expert analyses to gain a well-rounded understanding. If you have other questions or need assistance with less sensitive topics, feel free to ask!’

China’s government is restrictive of topics which can be discussed. RedNote, a Chinese a social media app which swept to prominence in the west this month, restricts politics entirely with users claiming they have been muted or banned for saying anything mildly controversial.

How is DeepSeek so cheap?

The company that makes it claims to have been able to do so with much, much less money and less computing power. It had to find a way with less, given that the US government restricted exports of the powerful semiconductor chips which have generally been used to power advanced AI.

This has led to questions about why US companies are raising billions and using so much energy, if similar can be achieved without it.

The worst hit financially was Nvidia, which was the world’s most valuable company… until yesterday.

After the world woke up to DeepSeek, the computer chip maker saw close to $600 billion wiped off its stock value, the biggest ever fall in a single day.

DeepSeek developers claim their R1 model was trained for around six million dollars (£4.8 million).

Anything even coming close to that figure would still be far less than the billions of dollars being spent by US firms, such as ChatGPT maker OpenAI, which is said to have spent five billion dollars (£4 billion) last year alone.

Start-up company DeepSeek was founded only recently in 2023 in Hangzhou, China, by hedge fund founder Liang Wenfeng.

Until now, the expenses quoted for AI development had only been getting more and more eye popping, with the UK announcing £14 billion of invesment days before the US announced £500 billion for the Stargate Project.

The AI industry, a huge disruptor to life as we know it, is now getting disrupted itself.

Dr Richard Whittle, an economist at the University of Salford, said: ‘The economics are quite simple, if a reasoning model can be developed so cheaply, then where is the value in the existing giants such as OpenAI, who have heavily invested in – what could be revealed to be – inefficient development models.’

He expects to see a ‘rethink’ in the market, with adoption potentially speeding up if companies can access AI more cheaply.

‘For the UK, DeepSeek raises interesting possibilities,’ he said. More efficient development methods may help with this developing sovereign AI capabilities.

‘However, the major issues facing AI in the UK are poor fundamental infrastructures and a declining talent base which limit our participation in the race,’ he said.

‘DeepSeek may well have sped everyone up, but not equally. The ability to capitalise on these developments requires considerable spending and strategic planning.’

Downing Street said the development demonstrates why the UK must ‘go further and faster to remove barriers to innovation’ in the AI sector.

Asked whether the Government would be open to using Chinese AI such as DeepSeek in Whitehall, the Prime Minister’s official spokesman said: ‘We’ve got some of the strongest data protection laws in the world and we will always ensure personal data and the operation of public services is handled responsibly and securely.

‘But more broadly on the developments over the last few days, the rapid development and breakthrough of new AI models demonstrates exactly why the UK is so focused on AI and why we need to go further and faster to remove barriers to innovation here to make the UK a more competitive market.

‘And whilst we’ve already got the third largest AI market in the world, we’ve got an opportunity to get ahead and do more, and that’s what our AI Opportunities action plan is all about.’

The spokesman added he would not ‘get ahead of specific models’ when asked whether he would rule out using Chinese AI in Whitehall.

Get in touch with our news team by emailing us at webnews@metro.co.uk.

For more stories like this, check our news page.