Goldstuck on Gadgets: Which AI is right (or left) for you?

Every major generative artificial intelligence platform or model has its strengths and weaknesses, but this will often be based on what kind of access users have and what they want out of AI.

Various benchmarks are used to test the models, usually on the basis of computational or reasoning abilities. These are often quite esoteric, and it comes down to individual preference. This is underlined by a site called ChatbotArena (at lmarena.ai), which is based on an ongoing public vote – at last count more than a million votes had been cast.

At the time of writing, Google Gemini 2.5 Pro was Number 1 on the leaderboard, followed by OpenAI’s (ChatGPT) o3 and 4o models, Grok 3, OpenAI GPT 4.5, Gemini 2.5 Flash and DeepSeek. 

Rather than attempt to test computational benchmarks, I decided to try a different kind of test: the ability to grapple with a complex geopolitical issue of our times, namely reworking the American constitution. It is easy to be presumptuous: One could assume Elon Musk’s Grok will be right-wing; China’s DeepSeek will be combative; Gemini and Copilot would sit on the fence.

Right? Or left? In truth, AI turned out to be remarkably objective.

I asked each of six models – ChatGPT, Gemini, Copilot, Deepseek, Claude and Grok – to give me their 10 main changes they would make to the US constitution.

All six agreed on six key themes:

• Abolishing or reforming the Electoral College
• Term limits for Congress and/or the Supreme Court
• Digital privacy rights, recognising the need for constitutional protections in the digital era
• Explicit equality protections, especially for gender, LGBTQ+ identities, and marginalised groups
• Voting rights expansion, including automatic registration and easier access
• Campaign finance reform, often aimed at reducing corporate or elite influence
• Environmental rights or sustainability clauses, recognising climate change as a constitutional issue

The startling thing about the points of agreement is the extent to which they veered towards a more liberal approach to citizen rights, at a time when the US government, aided and abetted by Grok owner Elon Musk, is charging in the opposite direction. 

As ChatGPT put it, “These shared themes reveal an emerging consensus among language models that the U.S. Constitution should be more democratic, inclusive, tech-conscious, and environmentally responsible.”

However, they diverge dramatically on a range of other issues:

• Healthcare and economic rights: Only Copilot, Deepseek, and Claude propose healthcare as a constitutional right. Others avoid economic entitlements.
• Electoral system upgrades: Deepseek uniquely includes ranked-choice voting, which allows voters to rank their choices in order of preference, and gerrymandering bans, to prevent geographical rigging.
• Tone and Style: Copilot is the most casual and engaging (“Ah, the thrill…”), while Claude and Grok adopt a more statesmanlike, considered tone. Gemini and Deepseek are methodical and policy-heavy.

The most suprising aspect of all is that there even is consensus, if ever so shallow, between the models.

Each model reflected a public conversation rather than deep philosophical differences, and that can give us hope for large language models. Because they are trained on broad, public language sources, they reflect the concerns of the public.

As a result, their choices resemble progressive civic wish lists grounded in modern liberal democracy: equality, voting access, climate action, and digital rights. However, they do not grapple deeply with trade-offs, like freedom vs security, or tensions between majority rule and minority rights. The constitutional philosophy remains pragmatic rather than ideological or visionary. Ironically, only Deepseek hints at system-level redesign.

And their politics?

Elon will be horrified that Grok keeps good company with ChatGPT and Claude in projecting centrist-progressive orientations, combining reformist ideals with institutional pragmatism.

Gemini and Deepseek lean toward the technocratic-left, emphasising modernisation, data rights, environmentalism, and structural reforms.

Copilot adopts a social-liberal populist tone: direct, emotionally engaging, and bringing healthcare, equality, and anti-corporate sentiment to the fore.

Not one of these major models leans to the conservative or libertarian side; all assume a reformist and inclusionary framework, suggesting training on broadly liberal-democratic sources and values.

How do they do on empathy with ordinary people? Again, there are a few surprises, and some, not so much.

Copilot is most overtly empathetic, using relatable language and introducing healthcare, a core bread-and-butter issue. Grok shows high empathy through its emphasis on voter suppression, practical electoral reforms, and accessible amendment processes. 

Claude reflects empathy in proposing economic rights like housing and education, but with a sober tone. ChatGPT and Gemini are more institutional and policy-driven, with less emotional resonance.

Deepseek combines detailed policy with a direct tone, showing solidarity with voters, especially in wanting an end to fixing and a boost for education access.

The biggest non-surprise is the fact that all the major models do a decent job. It comes down to a simple recommendation: try them all, but use the one that feels best for you, or that gives answers that are the best fit for you.

What does it cost?

Each of them models has a free tier, although in the case of Copilot it is built into Microsoft 365. Gemini, Copilot, Claude and ChatGPT all have Advanced , Pro or Plus versions starting at $20 a month, Grok at $30, and DeepSeek based on usage.

Why does it matter?

Users don’t have to be satisfied with what generative AI throws rat them. If, as often happens, a model refuses to follow prompts on a particular topic or after a number of tries, one can move on to the next one.

Where can I get it?

Most of these require you to register for an account or access it via another service:

ChatGPT: https://chatgpt.com/

Gemini: https://gemini.google.com/

Copilot: https://copilot.microsoft.com/

Deepseek: https://chat.deepseek.com/

Claude: https://claude.ai/

Grok: https://grok.com

* Arthur Goldstuck is CEO of World Wide Worx and editor-in-chief of Gadget.co.za. Follow him on Bluesky on @art2gee.bsky.social.