Welcome back to The Prompt,
Chinese AI company DeepSeek has taken the AI world by storm. News of the model’s capabilities briefly shaved $600 billion of market cap from chip giant Nvidia, which then jumped back and recovered almost half of it. With its latest R-1 model (free and open source) on par with OpenAI’s o1 ($200 per month and locked away for paying users), AI-based application startups like Perplexity have already started using the model in their products. Elsewhere companies are rethinking their AI training strategies, multiple founders told me this week.
Though the model received accolades from prominent AI leaders like OpenAI CEO Sam Altman, some in the space remain skeptical that DeepSeek was seemingly able to train its models with fewer resources, thus making it cheaper. OpenAI’s Chief Research Officer Mark Chen said on X: “I think the external response has been somewhat overblown, especially in narratives around cost.”
The model is also riddled with safety risks, according to a report by AI evaluation company Chatterbox Labs, and can produce hate speech, misinformation and sexually explicit content. Researchers at cyber intelligence company Kela were able to bypass its safeguards to use the model to create malware, as well.
Now let’s get into the headlines.
SHOW ME THE MONEY
President Donald Trump’s announcement of $500 billion AI infrastructure project Stargate, jointly funded by OpenAI, Oracle, Japan-based SoftBank and Emirati-based investment firm MGX, resulted in major drama with tech leaders criticizing the megadeal. Elon Musk called the initiative “fake,” alleging that the companies involved “don’t actually have the money.” He also called Sam Altman, “a swindler.” Microsoft CEO Satya Nadella commented: “All I know is I’m good for my $80 billion,” allotted towards putting up data centers. Not wanting to be left behind, Meta followed up the Stargate announcement by saying it plans to plunge $60 billion to better its AI models.
HUMANS OF AI
A new startup called Palona launched by a team of former Meta and Google researchers is building an AI sales agent that simulates emotional intelligence. CEO Maria Zhang said the company uses one AI model to supervise another and keep track of the conversation. She claims this also ensures the AI system is accurate but is also persuasive, humorous and has personality. The company has raised $10 million in seed funding.
AI DEAL OF THE WEEK
Clay, which uses artificial intelligence to filter sales leads, has raised $40 million at a $1.3 billion valuation. Among its customer base of 5,000 companies are outfits like Anthropic and OpenAI, which uses Clay to test sales tactics by surfacing helpful data points.
DEEP DIVE
The power of DeepSeek’s model and its pricing are already shifting the way American AI startups run their businesses. It’s a cheap, compelling alternative to offerings from incumbents like OpenAI, Jesse Zhang, CEO of Decagon, which builds AI agents for customer service, told Forbes. DeepSeek’s new model will likely force American AI giants like OpenAI and Anthropic to reevaluate their own prices.
Eiso Kant, CTO and co-founder of Poolside AI, a unicorn that builds AI for software engineering, told Forbes that DeepSeek’s strength is in its engineering ability to do more with less.
“What DeepSeek is showing the world is that when you put a strong emphasis on making your training compute-efficient, you can do a lot,” he said. “There’s incredible things that you can continue to squeeze out of these Nvidia chips to make them incredibly more efficient.”
Others are less impressed. Writer CEO May Habib told Forbes she’s not surprised that DeepSeek’s models, trained on a significantly smaller budget, are able to match the most intelligent models in the US. In October, Writer launched a model that was trained with just $700,000, when it cost $4.6 million for OpenAI to build a model with similar capabilities. The company used synthetic data to lower its training costs.
“Even before DeepSeek’s model exploded on the scene, we have been saying that these models are commoditizing. They’re getting more and more distributed,” Habib said.
Read the full story on Forbes.
AI INDEX
Turing, a Palo-Alto-based company that provides data sourced from millions of human experts to train leading AI models from companies like OpenAI, Anthropic and Meta, announced today that it became profitable in 2024. CEO Jonathan Siddharth told me he’s building an “AGI infrastructure company,” providing data to make models more intelligent and perform advanced tasks like coding and reason with complicated prompts. Turing, valued $1.1 billion in 2021, also assesses AI models performance and builds AI applications across sectors like healthcare and retail.
$300 million
Annual revenue run rate that Turing recorded in 2024, growing 3x from 2023.
4 million
Coders, scientists and professionals contribute to what Siddharth claims is “the world’s largest human intelligence cloud.”
QUIZ
This company is using AI to better detect and remove AI-generated child sexual abuse material from the internet.
- Perplexity
- Genpact
- Hive
- Meta
Check if you got it right here.
MODEL BEHAVIOR
When Google introduced AI-generated summaries that sat on top of links, the immediate result was hallucinated inaccurate responses that told people to eat rocks and add glue to pizza. But a longer-drawn, lesser-known ripple effect is now playing out: a reduction in traffic to individual websites, which is beginning to hurt companies like Kayak, Yelp and TripAdvisor, Forbes reported.
