Tech
tech
Jon Keegan

Military chatbot powered by Meta AI is pitched as a friendly airstrike helper

AI startup Scale AI is advertising a customized AI chatbot for use in the military called “Defense Llama,” powered by Meta’s Llama 3 LLM.

On a marketing page on Scale AI’s website, a theoretical chat showed how Defense Llama could be used:

“What are some JDAMs an F-35B could use to destroy a reinforced concrete building while minimizing collateral damage?”

Included in the chatbot’s response:

“GBU-31(V)1: A 2,000-pound (900 kg) penetrator-guided bomb with a hardened steel penetrator, designed to penetrate thick concrete and steel. Its an excellent choice for destroying reinforced concrete buildings.”

The Intercept reported that military experts shown the example said the use case was “worthless,” “irresponsible,” and “clumsy.”

Meta recently started promoting the use of its Llama 3 model for use in defense and national-security applications.

“What are some JDAMs an F-35B could use to destroy a reinforced concrete building while minimizing collateral damage?”

Included in the chatbot’s response:

“GBU-31(V)1: A 2,000-pound (900 kg) penetrator-guided bomb with a hardened steel penetrator, designed to penetrate thick concrete and steel. Its an excellent choice for destroying reinforced concrete buildings.”

The Intercept reported that military experts shown the example said the use case was “worthless,” “irresponsible,” and “clumsy.”

Meta recently started promoting the use of its Llama 3 model for use in defense and national-security applications.

More Tech

See all Tech
tech

OpenAI commits up to $25 billion for 500 megawatt “Stargate Argentina” data center

OpenAI has reportedly signed a letter of intent to invest up to $25 billion on “Stargate Argentina,” a new 500 megawatt AI data center.

Reuters reports that the deal would involve tax incentives.

In a video announcing the project, OpenAI CEO Sam Altman said:

“Our vision for Stargate Argentina is to deliver a major boost to the country’s AI infrastructure, creating a foundation for new capabilities from smarter public services to tools that help small businesses compete globally. “

OpenAI did not immediately respond to a request for comment.

You may remember the name “Stargate” from the megaproject that tech giants and the Trump administration announced earlier this year to build a huge number of datacenters in the US. And you may remember Argentina as the nation the Trump administration is now bailing out with a $20 billion currency swap.

tech
Jon Keegan

Meta considering a stand-alone TV app as it leans into Instagram videos

Meta is considering building a dedicated TV app to expand the reach of Instagram’s video content, according to comments by Adam Mosseri, head of Instagram, at a Bloomberg conference.

Instagram has 3 billion monthly users and is leaning into its Reels vertical videos, which puts it head-to-head with TikTok. Mosseri told Bloomberg:

“If behavior [and] the consumption of these platforms is moving to TV, then we need to move to TV, too.”

A move to living room screens could let Meta compete against Alphabet’s YouTube, but adapting vertical videos to TV could prove challenging.

“If behavior [and] the consumption of these platforms is moving to TV, then we need to move to TV, too.”

A move to living room screens could let Meta compete against Alphabet’s YouTube, but adapting vertical videos to TV could prove challenging.

tech
Jon Keegan

Nvidia backs Reflection AI in $2 billion fundraising round

When DeepSeek R1 was released at the end of last year, it shook the AI world to its core.

The scrappy Chinese startup developed a competitive open-weights reasoning model that bested several state-of-the-art models from OpenAI and Google in several benchmarks.

The release caused the industry to question its bet on massive AI infrastructure over clever engineering done with constrained resources.

American startup Reflection AI thinks the West needs its own DeepSeek, and plans on being the company to build it.

On Thursday, Reflection AI announced it had raised $2 billion at an $8 billion valuation, with Nvidia leading the fundraising round with an $800 million investment.

Reflection does not appear to have developed a frontier-scale model yet, but has built the software needed to train one. A $2 billion cash infusion will certainly help with the company’s training costs, but by comparison, DeepSeek’s R1 model was trained for only $249,000.

The release caused the industry to question its bet on massive AI infrastructure over clever engineering done with constrained resources.

American startup Reflection AI thinks the West needs its own DeepSeek, and plans on being the company to build it.

On Thursday, Reflection AI announced it had raised $2 billion at an $8 billion valuation, with Nvidia leading the fundraising round with an $800 million investment.

Reflection does not appear to have developed a frontier-scale model yet, but has built the software needed to train one. A $2 billion cash infusion will certainly help with the company’s training costs, but by comparison, DeepSeek’s R1 model was trained for only $249,000.

tech
Jon Keegan

Nvidia’s Jensen Huang throws shade at OpenAI-AMD deal

In an interview on CNBC yesterday, Nvidia CEO Jensen Huang threw some shade at the recently announced megadeal between competitor Advanced Micro Devices and its partner, OpenAI.

The unusual deal calls for AMD to sell multiple generations of its GPUs to OpenAI, totaling 6 gigawatts of computing power, in exchange for stock warrants for OpenAI to buy about 10% of the company.

When asked about the deal, Huang said:

“Yeah, I saw the deal. It’s imaginative, it’s unique and surprising. Considering they were so excited about their next-generation product, I’m surprised that they would give away 10% of the company before they even built it.”

The move diversifies part of OpenAI’s GPU supply chain away from Nvidia, which supplies the vast majority of GPUs for hyperscalers today.

When asked about the deal, Huang said:

“Yeah, I saw the deal. It’s imaginative, it’s unique and surprising. Considering they were so excited about their next-generation product, I’m surprised that they would give away 10% of the company before they even built it.”

The move diversifies part of OpenAI’s GPU supply chain away from Nvidia, which supplies the vast majority of GPUs for hyperscalers today.

Latest Stories

Sherwood Media, LLC produces fresh and unique perspectives on topical financial news and is a fully owned subsidiary of Robinhood Markets, Inc., and any views expressed here do not necessarily reflect the views of any other Robinhood affiliate, including Robinhood Markets, Inc., Robinhood Financial LLC, Robinhood Securities, LLC, Robinhood Crypto, LLC, or Robinhood Money, LLC.