Eric Peterson — Writing

Gaining a Civic Voice with Large Language Models

Tue, 05 Aug 2025 15:30:00 GMT

It’s the year 2025 and it’s a strange time to be a knowledge worker. Zoomed all the way out, you see investors and the world’s largest technology companies collectively burning a trillion+ dollars on the LLM-fueled hunt for productivity gains, aimed at upending the global economy as we know it. Zoomed all the way in, you see LLMs regularly struggle to turn their finely-tuned, probabilistic semantic mappings into even the most basic, durable value.

While the jury is still out on macroeconomic revolution, there’s at least broad consensus that there’s a there there; we are just collectively, through socio-technical coercion, in the phase of finding the contours of that value.

I remain skeptical of the overall impact the AI boom will have on my profession, but I’ve at least been able to derive personal value via an increased ability to challenge power structures in my adoptive country. Let me explain.

Asymmetric language proficiency

For a variety of well-documented structural reasons (which can be boiled down to exceptional English fluency), it’s comparatively easy for English-speaking immigrants to get settled in Sweden, but significantly more difficult for them to organically progress at Swedish language acquisition. A pronounced, asymmetric proficiency between production and comprehension of Swedish is common.

Even after 5 years here, the relatively few opportunities I have to produce Swedish (both written and oral) leave me at a stress-free-but-still-basic level (perhaps B1 on the CEFR scale). By contrast, I consume a significant amount of Swedish media which, over time, has led to proportionally higher levels of reading and listening comprehension (closer to a C1 level).

And the media I consume is disheartening: like most European countries, Sweden finds itself enmeshed in a wave of anti-immigration and xenophobia. Sweden’s minority, “pro-business” center-right government coalition finds itself wholly dependent on a larger, far-right anti-immigrant party, regularly putting it in ironic conflict with pro-business interests. They’re also pushing through legislation that directly and negatively impacts me, my colleagues and (I believe by extension) the economy at large.

Although my B1-level Swedish production prevents me from effectively engaging with these issues in any meaningful way, LLMs enable me to parlay my C1-level comprehension into C1-level writing. Who would’ve thought that a product whose founding research was, for decades, motivated by and benchmarked on translation, would excel at translation tasks?

By drafting content in my native language first, then translating interactively with an LLM, I’ve reached audiences I would never have imagined possible: including editorial pieces in a Big 3 Swedish newspaper, as well as a formal response to the government itself.

LLMs and translation in practice

If you find yourself in a similar situation, I’ve collected a few notes on how to leverage LLMs for translation in the context of civic engagement (or, generally!).

Use the right model for the job

AI companies these days are iterating very rapidly on their foundational models, often releasing multiple models that are (at least marketed to be) specialized for different types and levels of tasks. Be sure you’re using a model that’s up to the task; try several. In my experience, the difference in quality of output is stark and immediately noticeable.

Context is king, and it’s your responsibility to provide it

While modern models can perform web searches before providing an initial response, you’ll get better results if you do the research and provide the model with that context yourself: everything from mundane details like document structure, formatting, and length, to your intended audience and their motivations, as well as your own goals, purpose, and tone. Every detail, even if you’re repeating information embedded in the structure of your original document, helps tune the model toward a more ideal output.

In more formal contexts, or where you want specific terms to be used in the translation, it can also be helpful to provide a short glossary as part of your context (e.g. specifying how “concur” or “reject” should be translated, or mirroring vocabulary used in source documents or statements, e.g. to delineate between colloquial and legal terms used for “residency,” etc).

Guide away from a tendency toward direct translation

I don’t know if it’s an artifact of the linguistic relationship between English and Swedish, or of the training data used by the models I was working with, but I found that, even with explicit instructions to prefer idiomatic structures and vocabulary, initial outputs would often include stilted clauses that mapped nearly 1:1 to the source material.

You’ll need to lean on your familiarity with the target language to identify unnatural sounding output and guide the model toward more natural constructions.

Mistranslations happen, regularly even

While it’s exceedingly rare for a model to output grammatically incorrect translations, I’ve found the majority of my time, especially in the beginning of the process, is spent identifying plausible-sounding but just plain inaccurate translations.

Iterate in phases: first read carefully and identify and work to resolve mistranslations, then focus on clarity and flow. You’re playing the role of a translation editor where your biggest strength is that you also authored the original document, but where your biggest limitation is your grasp of the target language.

Batch feedback, apply fixes manually

Before iterating, I find it best to copy the initial translation into a separate document where I can manually apply changes (and track/revert changes if necessary). As your session goes on, the quality of the output tends to deteriorate, so it’s best to keep yourself in the loop and in control of the final product.

When you iterate with the model, batch your feedback: for example, in your responses, number each of the questions or points you want to make related to the translation. Depending on the company/product/model, it may go into an editing mode where it makes direct changes, or it may respond point-by-point. Evaluate each of the proposed changes and apply them manually to your document as you see fit.

Evaluation step

After you’ve accumulated edits and feel satisfied with your translation, I find it helpful to feed the translation back into the model (in a new session), asking it to evaluate the translated document on its own merits in the context of your goals: is it likely to be well-received by your intended audience, does it have the tone you’re going for, etc., and if not, ask for suggestions to remediate.

For particularly important or sensitive documents, you should pass your document by a trusted native speaker (ideally, an editor).

There are no shortcuts

LLMs are not magic. The outputs they produce can be no better than the input, context, and guidance they are provided. I don’t think I would have been as successful without extensive drafting and editing in my native language. A fairly high-level of Swedish comprehension was also critical (LLMs are no substitute for language learning).

Even still, the value is clear: I was able to get my voice and ideas into the public discourse. Hopefully this inspires others in similar situations to have their voice heard.

Response to the inquiry into stricter requirements for acquiring Swedish citizenship

Tue, 01 Apr 2025 07:00:00 GMT

The following is a copy of a formal response to a government consultation regarding tightened requirements for acquiring Swedish citizenship. I authored this around March 24th, then solicited signatures from foreign-born members of the tech community in Sweden (as well as their native allies), before officially submitting the response to the government the following week.

Image courtesy Håkan Dahlström

Summary

We the undersigned 379 labor migrants in the tech sector oppose the stricter requirements proposed in the inquiry. If put into force, these requirements will have an immediate and detrimental effect on Sweden’s ability to attract and retain highly skilled labor, of which there is an ongoing shortage. We in particular oppose the extended residency requirement, additional knowledge requirements, as well as the introduction of a self-sufficiency requirement.

In the event the proposal is to move forward, we concur with the recommendation that transitional provisions be put in place for citizenship applications received before the law goes into force. We however recommend an alternative formulation of those transitional rules.

Length of residence for acquisition of Swedish citizenship (section 7.4.3)

To be a labor migrant anywhere is to live a life of uncertainty. Putting down roots in a country where your right to work and reside is conditional on factors outside of your control (e.g. company restructuring, market conditions, or changes to laws) carries high financial and personal risk. Citizenship is the only way to unconditionally retain the right of residency and work and to eliminate that risk.

Sweden’s current residency requirement for citizenship (5 years in the normal case) is among the lowest in the world and therefore an invaluable differentiator in the competition for skilled foreign labor. All other things being equal, a faster path to citizenship, and therefore stability, would be the deciding factor in where a worker chooses to move.

Any increase to the residency length would therefore harm Sweden’s attractiveness on the global labor market. The proposed 8 year residency length, which would bring Sweden in line with many of its peers, would erase that advantage entirely. Anything above 8 years (which would approach a quarter of a working life) would be devastating to Sweden’s attractiveness.

The length of residence required to obtain Swedish citizenship should therefore not be increased beyond 5 years.

Demonstrating knowledge of Swedish and Swedish society (section 9.2.2)

Being an immigrant is a high-stress endeavor: navigating new bureaucracies, leaving behind family and friends, and adjusting to a new environment all take their toll. Attaching language and knowledge requirements to key administrative milestones in an immigrant’s journey only adds to that stress.

Sweden is exceptional in that no language or other knowledge requirements are currently in place for permanent residency or citizenship. This is a rare benefit for prospective workers and therefore a strong differentiator in Sweden’s attractiveness to foreign skilled labor. This is especially the case given the reality that the majority of the roles we fill are performed in English.

We moved to Sweden to pursue full-time work; fitting in language and other knowledge acquisition between work and family time can be challenging. While some language and knowledge skills may ultimately prove useful for integration in the long term, the ability to pursue that knowledge at one’s own pace and according to one’s own needs and abilities is an undeniable benefit, especially when compared to peer countries.

Additional language and knowledge requirements should therefore not be added.

The requirement that foreigners can support themselves (section 10.4.2)

Immigrating to a country is one of the largest changes one can make in one’s life. In making that decision, many of us needed to take into consideration our partners’ and/or our family’s lives.

Partners of skilled labor migrants have traditionally found it challenging to break into the Swedish labor market, especially when their careers are outside the tech sector. While the proposed introduction of a self-sufficiency requirement at the reserve amount (förbehållsbelopp) would not directly impact us as individuals, a higher income requirement could affect our families.

When evaluating countries to immigrate to, the ability for all members of our families to follow the path to citizenship together cannot be underestimated as a benefit. Any possibility for our family members’ rights to differ from ours over time will introduce friction into the process of attracting foreign labor. At the very least, it would significantly bias the types of individuals who would be willing to move here.

While we believe the amount proposed for self-sufficiency is reasonable, we must nonetheless oppose its introduction, especially if any higher amount is considered.

Transitional provisions (section 17.2)

A significant contributor in the success of Sweden as a society is the consistently high trust placed in its institutions. Given the proposal aims to alter legal conditions which materially formed the basis for our decision to move to Sweden, doing so retroactively would substantially harm our relationship with the state. We therefore concur with the need to have transitional provisions in place if the proposal is to go into force.

At a minimum, those transitional provisions should reflect what has been proposed: that applications sent in prior to the law going into effect should be evaluated under the rules in place at the time of application.

An even higher degree of trust would be maintained if citizenship applications would instead be evaluated based on the rules which were in force at the beginning of legal Swedish residency.

Concluding remarks

Thank you for the opportunity to provide feedback on this proposal. This referral response is hereby signed by the following labor migrants employed within the tech sector in Sweden. In cases where an individual did not wish for their name to be made public, the name of a concurring Swedish resident has been added in their place.

(Signatures omitted on this reproduction of the document)

Software Engineering and the Social Contract

Thu, 11 Jul 2024 22:06:00 GMT

From time to time, I get the question “Why did you move from the USA to Sweden?” The one-liner I deploy in most situations is that I moved “in search of a better life.” While this isn’t an entirely untrue statement, it’s a simplification of a more complex set of motivations related to social responsibility and values that not everyone is interested in hearing in the moment. This is my attempt at articulating those underlying reasons.

I am a software engineer by trade, and it’s an interesting time to be in my line of work. Never before in history has such a large group of workers had such a high degree of mobility. Look at the top economies across the world and you’ll struggle to find a country that doesn’t list software engineering under its “in-demand skills” list for labor migration.

Unlike other high-demand roles like doctors, healthcare professionals, and engineers of other sorts, software engineers need no formal qualifications when landing a job: no certification or license from some nationally recognized standards body, no years-long specialized higher education program–in fact, increasingly no degree is needed at all. In competitive labor markets and especially at multinational organizations, software jobs are typically performed in English too, reducing language as a barrier to mobility for many.

Given strong global demand, compensation tends to be relatively high for software engineers. Nowhere is this more true than the United States of America, where even entry-level engineers can earn six figures, with more experienced engineers clearing three times that in total compensation (even more for exceptionally experienced engineers and for those working at the largest tech companies).

Put simply: an English-speaking software engineer with more than a few year’s experience could choose to work just about anywhere in the world and live comfortably.

This, to me, has profound implications in the context of the social contract—the general idea that individuals give up some freedoms in exchange for broader social benefits, legitimizing the state and norms governing that society in the process. For most, the idea of a social contract is something of a farce: how many people can say that they chose the society that they live in?

But to be able to choose where to live and work means that staying in the USA is itself a choice, one that implied that not only was I fine with the state of things around me, but that I was happy to benefit disproportionately from it. Even if I didn’t feel like I was personally signing off on the gun violence epidemic or the student debt crisis or mass incarceration or the opioid crisis (etc), cashing a big check every month felt at least like legitimizing the values that bred those societal ills.

It’s more than abstract philosophy too: you can vote with your wallet and avoid that unethical chicken sandwich chain or subscription service, but a software engineer can have several orders of magnitude more impact by voting with their labor (and redirecting the tax revenue that it generates elsewhere).

…Which is what I chose to do. Sweden is no perfect society, but the sum of the vectors of its social values are more closely aligned with my own than my home country’s. I had a good life there, just as I have a good life here; I wasn’t looking for a better life for myself, but instead a better way for all to live that I could contribute to.

Goodbye Seattle

Thu, 31 Dec 2020 14:00:00 GMT

For those not already aware, some personal news: I’ve moved to Stockholm, Sweden.

Given the new year, I wanted to briefly reflect on how I ended up here. 2020 has been a trying year for everyone. In one sense, I haven’t been impacted at all by everything going on in the world: I will end the year just as I began it, developing software in front of a computer screen from home.

In another sense, 2020 was the peak of a multi-year run-up in anxiety and mental fatigue for me. How little my good fortune has changed compared with the unimaginably many misfortunes around me became an unignorable cognitive dissonance. I spent a lot of energy worrying about systemic problems I struggle to see solutions to.

I believe my ideal social contract would be something like this: I want to be free to focus my energy on my work, relationships, and interests. For this freedom, I’d gladly contribute to a state that administers programs that reduce the need for individuals like me to have to focus their energy on anything else.

In practice, this is stuff like equal access to healthcare, education, childcare, welfare, and time-off: things that in the US are not a given and end up distributed unevenly.

So I find myself here, a guest of Sweden, which at least at first glance seems to operate under such a social contract. I’ve enjoyed learning the language, the culture, and look forward to a little more of a hands-on approach to it all in the coming year or so, pandemic permitting.

While uprooting your life and moving halfway around the world is not possible for everyone, I hope that those also suffering from the mental fog of the last year and beyond find time to reflect on why, and find a better allocation of energy in the coming year.