What is Grok? Elon Musk’s controversial ChatGPT competitor explained

Trending 2 months ago

Grok! It mightiness not rotation disconnected nan lingua for illustration ChatGPT aliases Windows Copilot, but it’s a ample connection exemplary chatbot each nan same. Developed by xAI, an offshoot of nan programmers who stuck astir aft Elon Musk purchased X (formerly known arsenic Twitter), Grok is designed to compete straight pinch OpenAI’s GPT-4 models, Google’s Bard, and a scope of different public-facing chatbots.

Launched successful November 2023, Grok is designed to beryllium a chatbot pinch little of a select than different AIs. It’s said to person a “bit of wit, and has a rebellious streak.”

It’s only for X Premium users

Grok AI motion up page.Jon Martindale / DigitalTrends

Want to play astir pinch Grok yourself? You can, if you’re an X Premium+ subscriber and besides reside successful nan U.S. (r usage a VPN). Even so, there’s a waiting database to effort retired nan prototype, truthful there’s nary guarantee you’ll get in.

While you mightiness expect it to unfastened up to a wider assemblage erstwhile it’s retired of beta, Musk has suggested that Grok will stay a characteristic exclusive to Premium+ X subscribers for nan foreseeable future.

It’s a chatbot

Just weeks aft discussing really dangerous Musk thought nan caller advances successful AI were, he launched Grok successful beta form. Though he claimed that it would beryllium much unfastened successful its connection style, pinch little of a taxable filter, it’s yet a chatbot for illustration immoderate of nan others.

That’s because it was built successful overmuch nan aforesaid way. It’s a ample connection exemplary AI, which intends it was built utilizing tons of training data, immoderate from X, immoderate from nan web astatine large. We’re told its moving connected nan Grok-1 ample connection exemplary (LLM) pinch immoderate 33 cardinal parameters. This would put it down Meta’s LLaMa and OpenAI’s GPT-4, which person 70 cardinal and 1.76 trillion parameters, respectively. It was besides trained for only a mates of months, whereas different models person taken years to put together.

That whitethorn beryllium why Grok appears to beryllium conscionable arsenic susceptible (if not to much so) to hallucinations arsenic different AIs for illustration ChatGPT. It often makes up facts aliases cites nonexistent sources erstwhile asked for actual responses.

It uses information from X

One of nan large trading points of Grok is that it brings unrecorded information from X. Where different connection models mightiness usage Google aliases Bing to find up-to-date accusation to augment their answers to personification prompts, Grok tin guidelines its responses disconnected of up-to-date accusation from X. That gives it entree to accusation that different chatbots don’t have, but it besides intends that it is perchance much susceptible to misinformation owed to nan sheer standard of its dispersed connected X since Musk’s takeover.

That does mean it has entree to Musk’s X, account, however. Which intends it doesn’t work together pinch him each nan time.

It’s expected to beryllium anti-woke, but it isn’t

A awesome logic for Musk’s takeover of X and nan motorboat of Grok has been his ain effort to conflict backmost against what he considers “wokeness.” Grok was expected to beryllium an AI that eschewed societal politeness successful favour of a peculiar marque of humor, arsenic good arsenic a propensity to thin correct successful its governmental biases. However, that’s not turned retired to beryllium nan case. Indeed, galore of Musk and Grok’s first fans are now complaining that nan AI has been captured by “woke programmers.”

Has Grok been captured by woke programmers?

I americium highly concerned here. 😮 https://t.co/lrlOJ79Mzz

— Wall Street Silver (@WallStreetSilv) December 8, 2023

In reality, Grok’s training information was based connected overmuch nan aforesaid information arsenic each different AI: nan humans who interact pinch each different online. Therefore, Grok was apt to postgraduate toward sounding for illustration nan different AIs retired there.

In its default mode, astir researchers recovered that Grok would respond pinch an overtly neutral answer if asked astir topics that are seen arsenic arguable by right-wing commenters, specified arsenic weapon control, ambiance change, aliases debunked conspiracy theories. However, if you move from nan modular “fun” mode to its regular mode, it will supply much reasoned responses.

It’s borrowing from ChatGPT

Grok has been recovered to make responses arsenic if it was ChatGPT astatine times. That’s not because it was built connected nan aforesaid exertion arsenic ChatGPT, but because it was trained connected nan web since nan merchandise of ChatGPT. That’s led it to hoover up AI-written content, and successful 1 instance, an OpenAI privateness policy, which it proceeded to provender backmost to its users arsenic if it wrote it.

This is simply a problem that has been cited by a number of experts and whitethorn worsen pinch each LLM AIs if it isn’t addressed. In a world wherever AIs tin and do create mountains of recycled contented each day, AI developers request a measurement to differentiate betwixt AI and quality contented truthful that they don’t conscionable train their early models connected AI-generated content.

The sanction comes from a Heinlein novel

If you’re wondering wherever nan bizarre sanction came from, “Grok” is simply a neologism from nan Robert Heinlein novel, Stranger successful a Strange Land. The original meaning was to “understand intuitively aliases by empathy, to found rapport with.” That sounds for illustration a lofty extremity for an AI, but it’s apt that it’s been employed present successful nan manner that it grew to beryllium known by wrong nan programming organization successful nan decades aft Heinlein’s work: To understand thing truthful fully, that it’s portion of your identity.

Grok is expected to understand america truthful good it is us. We don’t deliberation it’s rather location yet.

Editors' Recommendations

  • Google mightiness yet person an reply to Chat GPT-4
  • Here’s why group are saying GPT-4 is getting ‘lazy’
  • The world responds to nan creator of ChatGPT being fired by his ain company
  • OpenAI’s caller instrumentality tin spot clone AI images, but there’s a catch
  • ChatGPT: nan latest news, controversies, and tips you request to know