Meet Sol.

The cereal company and greenhouse motor demonstrations are not showing a narrow cereal-and-motor personality hidden inside the model. They show what happens when a general intelligence can combine research, calculation, design, writing, coding and tools. It can develop a market position, draft packaging copy, create a budget, produce a prototype website, reason about motor requirements and write a test plan.

It cannot physically climb into the greenhouse with a socket set. It can perform a remarkable amount of the intellectual work surrounding that job. Think less “improved Google” and more “small consulting firm trapped in your computer.”

Instead of asking, “Tell me about investing in apartment buildings,” give it the whole job:

That is not merely a question. It is an engagement letter for a temporary research team. You can tell it to stop after inspecting the sources, show its assumptions before calculating, or review the result as a spectacularly grumpy credit officer. The first output is a draft; the follow-up conversation is where the work becomes yours.

Investments: probabilities, not prophecy 📈🧙‍♂️🧃

Sol can compare asset classes, examine funds, explain financial statements, research current fees and holdings, model portfolios, calculate nominal and inflation-adjusted returns, run Monte Carlo simulations, compare tax scenarios and show how different assumptions alter the outcome.

Ask for ranges and sensitivities, not a vision of tomorrow’s closing price. If someone announces a 78.4 percent probability that one stock will rise next year, the decimal point may be wearing a fake moustache.

Ask it to argue against you when enthusiasm appears to be driving the bus. “Identify the strongest case against this investment, the evidence I may be discounting, and events that could create permanent rather than temporary loss” is often more useful than “Is this good?”

Ottawa property: paperwork meets weather 🏘️❄️🦩

Upload the listing, inspection report, floor plans, tax figures, insurance estimate, condo documents, rent roll, leases, mortgage proposal and repair information. Sol can extract the important numbers, find contradictions, build scenarios and produce a due-diligence checklist.

Require three categories: facts supported by supplied documents; current information verified from reliable sources; and assumptions still requiring confirmation. That prevents facts, guesses and realtor adjectives from becoming one casserole.

For a rental property, ask for interest-rate sensitivity, break-even occupancy, major-repair reserves, exit costs and a “what goes wrong in the first 24 months?” analysis. It cannot smell mould, inspect the foundation or discover the neighbour’s midnight tuba schedule. It can make sure you arrive at those human judgments with a much clearer picture.

Sitcom development: build a room, not a joke machine 📺🛋️🦞

Do not begin with “write a funny sitcom.” That often produces something with the spiritual atmosphere of an airline safety video. Start with the premise, tone, characters, setting, audience and comparable programs. Then examine the series engine: why must these people continue colliding every week?

Sol can help develop a logline, relationship map, series bible, pilot beats, episode premises, season arcs, scenes, a one-page pitch, a longer pitch document, a deck and a verbal presentation. Upload scenes whose tone works and scenes that feel false. Ask it to identify the structural and tonal difference without imitating another writer’s dialogue.

For current broadcaster requirements, Canadian funding structures, producer preferences or submission rules, ask for web research and current sources. Television pitching advice can age faster than milk in an Ottawa heat wave.

A comic website: “more sinister government pamphlet” 🕸️🪩🕵️

Describe the comic, audience, visual tone, colours, navigation and purpose. Sol can design and build a working webpage, write the HTML, CSS and JavaScript, create or request imagery, test the layout and revise it from your reaction.

You do not need to know whether the problem is line height, grid proportions or a haunted CSS margin. “This looks like a venture-capital firm selling powdered meals” is legitimate design feedback. Sol can translate the reaction into a technical change.

Guitar practice: design for the human who exists 🎸⏱️🥨

Tell it the player’s ability, musical taste, instruments, limitations, attention span, objectives and realistic schedule. Upload lesson notes, chord sheets, song lists, recordings and a teacher’s observations.

After a week, report what actually happened. If he skipped scales, enjoyed rhythm work and still stumbles at bar 12, the model can revise the plan without defending its original schedule like constitutional law.

Workouts: evidence, constraints and the peculiar shoulder 🏋️‍♂️🧠🦆

Supply goals, age, training history, equipment, schedule, injuries, preferences and anything a doctor or physiotherapist has said. Ask it to review weekly volume, movement balance, fatigue, progression and recovery. Most importantly, tell it which exercises you hate; the scientifically exquisite program you abandon is inferior to the good program you perform.

For pain, disease, medications or rehabilitation, use it to organize information and prepare professional questions. An AI-generated table is not a writ from Zeus.

Baldness: the gravest frontier 🥚🔭🌴

Sol can research current treatments, compare evidence, explain mechanisms, list risks, locate trials and help prepare questions for a dermatologist. It cannot bully the follicles into production through superior reasoning effort.

A useful medical review should clarify the kind and cause of hair loss, distinguish approved and off-label treatments, compare transplantation and cosmetic approaches, identify experimental research and avoid treating influencer testimonials as clinical evidence. The machine is ready; the follicles remain noncommittal.

Yes, you can ask Sol to prepare a generic surety facultative reinsurance agreement from supplied parameters. A better approach is to provide authorized, non-confidential examples and ask it to identify recurring structures before drafting.

It can work through definitions, scope of cover, attachment, limits, exclusions, premium, claims control, settlements, notice, records, audit, recoveries, subrogation, errors and omissions, offset, insolvency, governing law, arbitration, confidentiality, sanctions, privacy and termination. Ordinary civilians may perish somewhere around “claims cooperation,” but Sol remains disturbingly cheerful.

For comparison work, ask for changes in risk allocation, claims authority, notice, settlement control, insolvency exposure, dispute resolution, termination rights and definitions. Tell it to ignore stylistic edits unless they introduce ambiguity.

For experiments, use synthetic facts, public documents, generalized parameters or templates you are permitted to use. Incorporating AI into actual legal workflows should be an organizational decision involving the appropriate legal, privacy, security and IT people.

Use High when the task contains several stages, conflicting evidence, many documents, difficult tradeoffs, significant calculations or a meaningful risk of missing something. It takes longer and uses more of the Plus allowance; that expense is worthwhile when the result is valuable.

Leave it at the normal setting for rewriting a paragraph, explaining a concept, generating twenty premises, making a short guitar drill or deciding whether “Baldness Avenger” is a strong superhero name. It is, incidentally, not.

If you begin using the allowance quickly, Luna can handle bulk extraction and reformatting; Terra can handle everyday professional work; Sol on High can be reserved for the jobs that genuinely deserve it.

State what you want produced, why you need it, who it is for, which facts and constraints matter, what sources it should use, what a good result looks like, what it must not assume and what should be checked before finishing.

You can also say, “I have not thought this through sufficiently. Ask me one question at a time until you understand the goal, then recommend an approach.” That is powerful because the model can help identify the question you did not know you needed to ask.

The machine can still be wrong 🥸📉

Sol can misread an instruction, overlook a clause, perform an incorrect calculation, misunderstand a source, cite something irrelevant or fill a gap with a plausible invention. Better models do this less often; “less often” and “never” live in different provinces.

For consequential work, expose assumptions, request sources, show calculations, identify uncertainty and check the result against the original materials. Use it to improve your judgment, not to abolish the fact that judgment is required.

It cannot promise returns, conduct a physical inspection, make an unsafe engineering design safe through eloquence, diagnose disease, or turn a mediocre sitcom premise into thirty years of residuals without writers, performers, producers, schedules, budgets and the mysterious intervention of luck.