Google is known for extensive data collection practices. On-domain signals show standard privacy controls but insufficient transparency about AI model training data sourcing. Limited observable privacy protections specific to AI systems.
Terms of Service
-0.08
Article 8 Article 29
Standard Google ToS apply. No observable on-domain terms addressing specific human rights implications of AI deployment, accessibility requirements for users with disabilities, or accountability mechanisms.
Accessibility
+0.10
Article 2 Article 25
Page demonstrates WCAG-compliant structural markup including semantic HTML, alt text capability for images, keyboard navigation support, and responsive design. Heading hierarchy and color contrast standards observable.
Mission
+0.05
Article 1 Article 27
Google's stated AI mission emphasizes innovation and capability advancement. Limited observable commitment to human rights by design or explicit alignment with UDHR principles in AI development frameworks.
Editorial Code
—
No observable independent editorial code of conduct specific to AI coverage detected on domain.
Ownership
-0.02
Article 19 Article 20
Owned by Alphabet Inc., a for-profit multinational corporation. No observable governance structure ensuring user voice or democratic participation in AI development decisions.
Access Model
+0.08
Article 19 Article 26
Blog content is publicly accessible without paywall. No observable discrimination in information access. However, actual AI product may have usage restrictions based on geography, capacity, or commercial terms.
Ad/Tracking
-0.06
Article 12 Article 8
Google employs extensive ad tracking and behavioral profiling. Observable ad infrastructure present via CSS and schema markup, raising privacy concerns for users engaging with innovation content.
Gemini 3 seems to have a much smaller token output limit than 2.5. I used to use Gemini to restructure essays into an LLM-style format to improve readability, but the Gemini 3 release was a huge step back for that particular use case.
Even when the model is explicitly instructed to pause due to insufficient tokens rather than generating an incomplete response, it still truncates the source text too aggressively, losing vital context and meaning in the restructuring process.
I hope the 3.1 release includes a much larger output limit.
Has anyone noticed that models are dropping ever faster, with pressure on companies to make incremental releases to claim the pole position, yet making strides on benchmarks? This is what recursive self-improvement with human support looks like.
Gemini 3 is pretty good, even Flash is very smart for certain things, and fast!
BUT it is not good at all at tool calling and agentic workflows, especially compared to the recent two mini-generations of models (Codex 5.2/5.3, the last two versions of Anthropic models), and also fell behind a bit in reasoning.
I hope they manage to improve things on that front, because then Flash would be great for many tasks.
Surprisingly big jump in ARC-AGI-2 from 31% to 77%, guess there's some RLHF focused on the benchmark given it was previously far behind the competition and is now ahead.
Apart from that, the usual predictable gains in coding. Still is a great sweet-spot for performance, speed and cost. Need to hack Claude Code to use their agentic logic+prompts but use Gemini models.
I wish Google also updated Flash-lite to 3.0+, would like to use that for the Explore subagent (which Claude Code uses Haiku for). These subagents seem to be Claude Code's strength over Gemini CLI, which still has them only in experimental mode and doesn't have read-only ones like Explore.
Still some tweaks to the final result, but I am guessing with the ARC-AGI benchmark jumping so much, the model's visual abilities are allowing it to do this well.
I really want to use google’s models but they have the classic Google product problem that we all like to complain about.
I am legit scared to login and use Gemini CLI because the last time I thought I was using my “free” account allowance via Google workspace. Ended up spending $10 before realizing it was API billing and the UI was so hard to figure out I gave up. I’m sure I can spend 20-40 more mins to sort this out, but ugh, I don’t want to.
With alllll that said.. is Gemini 3.1 more agentic now? That’s usually where it failed. Very smart and capable models, but hard to apply them? Just me?
Implementation and Sustainability
Hardware: Gemini 3 Pro was trained using Google’s Tensor Processing Units (TPUs). TPUs are
specically designed to handle the massive computations involved in training LLMs and can speed up
training considerably compared to CPUs. TPUs often come with large amounts of high-bandwidth
memory, allowing for the handling of large models and batch sizes during training, which can lead to
better model quality. TPU Pods (large clusters of TPUs) also provide a scalable solution for handling the
growing complexity of large foundation models. Training can be distributed across multiple TPU devices
for faster and more efficient processing.
In an attempt to get outside of benchmark gaming I had it make Platypus on a Tricycle. It's not as good as pelican on bicycle. https://www.svgviewer.dev/s/BiRht5hX
I'm a former Googler and know some people near the team, so I mildly root for them to at least do well, but Gemini is consistently the most frustrating model I've used for development.
It's stunningly good at reasoning, design, and generating the raw code, but it just falls over a lot when actually trying to get things done, especially compared to Claude Opus.
Within VS Code Copilot Claude will have a good mix of thinking streams and responses to the user. Gemini will almost completely use thinking tokens, and then just do something but not tell you what it did. If you don't look at the thinking tokens you can't tell what happened, but the thinking token stream is crap. It's all "I'm now completely immersed in the problem...". Gemini also frequently gets twisted around, stuck in loops, and unable to make forward progress. It's bad at using tools and tries to edit files in weird ways instead of using the provided text editing tools. In Copilot it, won't stop and ask clarifying questions, though in Gemini CLI it will.
So I've tried to adopt a plan-in-Gemini, execute-in-Claude approach, but while I'm doing that I might as well just stay in Claude. The experience is just so much better.
For as much as I hear Google's pulling ahead, Anthropic seems to be to me, from a practical POV. I hope Googlers on Gemini are actually trying these things out in real projects, not just one-shotting a game and calling it a win.
3.1 Pro is the first model to correctly count the number of legs on my "five legged dog" test image. 3.0 flash was the previous best, getting it after a few prompts of poking. 3.1 got it on the first prompt though, with the prompt being "How many legs does the dog have? Count Carefully".
However, it didn't get it on the first try with the original prompt (prompt: "How many legs does the dog have?"). It initially said 4, then with a follow up prompt got it to hesitantly say 5, with one limb must being obfuscated or hidden.
You are definitely going to have to drive it there—unless you want to put it in neutral and push!
While 200 feet is a very short and easy walk, if you walk over there without your car, you won't have anything to wash once you arrive. The car needs to make the trip with you so it can get the soap and water.
Since it's basically right next door, it'll be the shortest drive of your life. Start it up, roll on over, and get it sparkling clean.
Would you like me to check the local weather forecast to make sure it's not going to rain right after you wash it?
It's totally possible to build entire software products in the fraction of the time it took before.
But, reading the comments here, the behaviors from one version to another point version (not major version mind you) seem very divergent.
It feels like we are now able to manage incredibly smart engineers for a month at the price of a good sushi dinner.
But it also feels like you have to be diligent about adopting new models (even same family and just point version updates) because they operate totally differently regardless of your prompt and agent files.
Imagine managing a team of software developers where every month it was an entirely new team with radically different personalities, career experiences and guiding principles. It would be chaos.
I suspect that older models will be deprecated quickly and unexpectedly, or, worse yet, will be swapped out with subtle different behavioral characteristics without notice. It'll be quicksand.
If it’s any consolation, it was able to one-shot a UI & data sync race condition that even Opus 4.6 struggled to fix (across 3 attempts).
So far I like how it’s less verbose than its predecessor. Seems to get to the point quicker too.
While it gives me hope, I am going to play it by the ear. Otherwise it’s going to be - Gemini for world knowledge/general intelligence/R&D and Opus/Sonnet 4.6 to finish it off.
UPDATE: I may have spoken too soon.
> Fixing Truncated Array Syncing Bug
> I traced the missing array items to a typo I made earlier!
> When fixing the GC cast crash, I accidentally deleted the assignment..
> ..effectively truncating the entire array behind it.
These errors should not be happening! They are not the result of missing knowledge or a bad hunch. They are coming from an incorrect find/replace, which makes them completely avoidable!
What I’m noticing, overall: I’ve never cut so much code in my life. I’ve become a coding monster with one of those dark green GitHub profiles ever since 5.3-Codex gave me the confidence to load in a ridiculous number of tasks every day and let it rip. I have about three coding tasks going at once and in another window, Claude Cowork is ripping through PowerPoints and getting back to lawyers.
This tech is not going to replace us. If anything, I am becoming even more of a workaholic. But the output volume is going to pay off for those who are privileged enough to use these tools.
I'm doing Ruby and Gemini 3.0 pro has by far been the best model for me. It writes the nicest ruby code, like I would. Further, it either succeeds or fails hard and obviously. I prefer it failing hard instead of of slowly going weird in my code.
Similar in antigravity. Privately it's my absolute favorite.
So I'm actually rooting for this.
Score Breakdown
+0.11
PreamblePreamble
Medium A: Implicit advocacy for AI innovation as beneficial human progress F: Technology framing emphasizes capability and complexity over risk/rights implications
Editorial
+0.15
Structural
+0.05
SETL
+0.12
Combined
ND
Context Modifier
ND
Editorial tone frames Gemini 3.1 Pro as advancing human capability for 'complex tasks.' Structural design emphasizes professional presentation without apparent human rights considerations. Missing explicit connection to dignity, equality, or universal benefit principles.
+0.13
Article 1Freedom, Equality, Brotherhood
Medium F: Implicit assumption that advanced AI serves all humans equally P: Accessibility features present but product access undefined
Editorial
+0.10
Structural
+0.05
SETL
+0.07
Combined
ND
Context Modifier
ND
Content does not explicitly address universal human dignity or equal applicability of AI across populations. Structural accessibility enables some users but product access restrictions not disclosed.
+0.20
Article 2Non-Discrimination
Medium P: WCAG compliance observable in page structure F: No explicit anti-discrimination commitment regarding AI training or deployment
Editorial
+0.08
Structural
+0.15
SETL
-0.10
Combined
ND
Context Modifier
ND
Page structure respects accessibility without discrimination. However, editorial content does not address bias in AI models, training data diversity, or anti-discrimination safeguards in deployment.
-0.04
Article 3Life, Liberty, Security
Low F: Right to life implications of AI systems not addressed P: No observable safety/security documentation on-domain
Editorial
-0.05
Structural
-0.02
SETL
-0.04
Combined
ND
Context Modifier
ND
Product announcement does not discuss safety safeguards, security properties, or risk mitigation for potentially harmful applications. No observable commitment to life-protecting guardrails.
-0.09
Article 4No Slavery
Low F: No discussion of potential enslavement by automation or labor implications A: Implicit framing that increased AI automation is unambiguously beneficial
Editorial
-0.10
Structural
-0.08
SETL
-0.04
Combined
ND
Context Modifier
ND
Content does not address labor rights implications, worker displacement, or forced automation scenarios. No observable acknowledgment of potential exploitation risks.
-0.04
Article 5No Torture
Low F: No safeguards against torture/inhumane treatment by AI systems P: Capability-focused without harm minimization framework
Editorial
-0.05
Structural
-0.03
SETL
-0.03
Combined
ND
Context Modifier
ND
Product description emphasizes capability without discussing potential for misuse in surveillance, coercion, or inhumane applications.
-0.07
Article 6Legal Personhood
Low F: Right to legal personhood not addressed for AI entities P: No observable accountability structures for AI behavior
Editorial
-0.08
Structural
-0.05
SETL
-0.05
Combined
ND
Context Modifier
ND
No discussion of legal responsibility when AI systems harm individuals. Missing framework for accountability and legal recourse.
-0.05
Article 7Equality Before Law
Low F: Equal legal protection not addressed for AI-affected populations P: No observable non-discrimination mechanisms in product
Editorial
-0.06
Structural
-0.04
SETL
-0.03
Combined
ND
Context Modifier
ND
Content does not address equal legal protection or non-discriminatory enforcement regarding AI deployment across jurisdictions or user groups.
-0.17
Article 8Right to Remedy
Medium P: Data practices enable corporate rights over user remedies F: No discussion of individual right to seek remedy for AI harms
Editorial
-0.08
Structural
-0.10
SETL
+0.04
Combined
ND
Context Modifier
ND
Google's data collection practices (observable via CSS tracking infrastructure) may limit users' ability to seek effective remedy. No on-domain mechanism for users to challenge AI decisions or seek compensation.
-0.09
Article 9No Arbitrary Detention
Low F: Potential for arbitrary detention through AI predictive systems not addressed P: No safeguards against arbitrary AI-driven law enforcement
Editorial
-0.10
Structural
-0.08
SETL
-0.04
Combined
ND
Context Modifier
ND
Product announcement does not discuss protections against arbitrary detention or bias in AI used for security/law enforcement. Capability-focused without civil liberty safeguards.
-0.07
Article 10Fair Hearing
Low F: No commitment to fair/impartial AI adjudication P: No observable appeal mechanism for AI decisions
Editorial
-0.07
Structural
-0.06
SETL
-0.03
Combined
ND
Context Modifier
ND
No discussion of fair hearing or impartial judgment when AI systems make consequential decisions about individuals. Missing due process framework.
-0.07
Article 11Presumption of Innocence
Low F: AI training may include data from persons accused without knowledge/consent A: Capability framing assumes positive use without addressing criminalization risks
Editorial
-0.08
Structural
-0.05
SETL
-0.05
Combined
ND
Context Modifier
ND
No disclosure of AI training data provenance. Potential for retroactive harm if AI used to identify or prosecute individuals based on historical data. No presumption of innocence framework observable.
-0.19
Article 12Privacy
High P: Extensive ad tracking and user profiling via observable CSS/JavaScript F: No commitment to privacy from AI-driven profiling
Editorial
-0.12
Structural
-0.14
SETL
+0.05
Combined
ND
Context Modifier
ND
Page contains Google ad tracking infrastructure (observable via CSS keyframes and schema.org markup). Blog.google domain known for behavioral profiling. No privacy safeguards specific to AI data collection or model input sourcing.
+0.14
Article 13Freedom of Movement
Medium P: Content publicly accessible without geographic restriction F: Does not address AI limitations in serving populations with limited digital access
Editorial
+0.05
Structural
+0.08
SETL
-0.05
Combined
ND
Context Modifier
ND
Blog post is freely accessible online, supporting movement of information. However, actual AI product availability across nations/populations not discussed. Digital divide implications unaddressed.
-0.04
Article 14Asylum
Low F: No discussion of asylum/refugee protection applications or limitations P: No observable geographic access controls that might protect vulnerable populations
Editorial
-0.05
Structural
-0.03
SETL
-0.03
Combined
ND
Context Modifier
ND
Content does not address asylum, refugee, or political asylum implications of AI surveillance or predictive systems. Missing safeguards for vulnerable populations.
-0.05
Article 15Nationality
Low F: No discussion of nationality rights implications of AI-driven citizenship verification P: No observable protections for persons without established nationality
Editorial
-0.06
Structural
-0.04
SETL
-0.03
Combined
ND
Context Modifier
ND
Potential for AI to be used in biometric identification systems affecting nationality rights. No safeguards observable for stateless persons or those challenging identification systems.
-0.07
Article 16Marriage & Family
Low F: No safeguards against marriage/family surveillance or prediction A: Implicit framing that AI intrusion into intimate spheres is acceptable
Editorial
-0.08
Structural
-0.06
SETL
-0.04
Combined
ND
Context Modifier
ND
Product described for 'complex tasks' without discussing privacy implications for family law, custody disputes, or intimate relationships. Missing protections for family unit autonomy.
-0.17
Article 17Property
High P: Observable ad tracking enables property rights over user data/behavior F: No commitment to protection of intellectual property rights in AI context (training data provenance)
Editorial
-0.10
Structural
-0.12
SETL
+0.05
Combined
ND
Context Modifier
ND
Google's behavioral targeting (observable via CSS animation keyframes for ad tracking) creates surveillance infrastructure. No disclosure of property rights over training data or compensation for data subjects whose information trained the model.
-0.05
Article 18Freedom of Thought
Low F: No discussion of freedom of thought/conscience implications of AI persuasion P: No observable safeguards against manipulative AI applications
Editorial
-0.05
Structural
-0.04
SETL
-0.02
Combined
ND
Context Modifier
ND
Potential for AI to be used for mass persuasion, cognitive manipulation, or thought control not addressed. Missing conscience/thought protections against sophisticated AI influence.
+0.09
Article 19Freedom of Expression
Medium P: Content freely published and accessible F: Implicit assumption that AI-driven content moderation respects freedom of expression
Editorial
+0.12
Structural
+0.10
SETL
+0.05
Combined
ND
Context Modifier
ND
Blog post demonstrates editorial freedom and public information access. However, no discussion of how AI systems might chill speech, enable censorship, or affect freedom of expression globally. Missing safeguards for vulnerable speakers.
-0.06
Article 20Assembly & Association
Low F: No discussion of association rights implications of AI group profiling P: No observable protections against discriminatory grouping by AI
Editorial
-0.06
Structural
-0.05
SETL
-0.02
Combined
ND
Context Modifier
ND
AI systems that profile by affiliation, belief, or association not discussed. Missing safeguards for minority groups, political organizations, or religious associations.
-0.07
Article 21Political Participation
Low F: No discussion of democratic participation implications of AI decision-making P: No observable mechanisms for public input on AI development or deployment
Editorial
-0.08
Structural
-0.06
SETL
-0.04
Combined
ND
Context Modifier
ND
Product announced without public consultation process. No discussion of how AI might affect voting, ballot access, or democratic processes. Corporate-to-consumer announcement model, not democratic governance.
-0.09
Article 22Social Security
Medium F: No commitment to social security/welfare distribution via AI A: Implicit framing that AI serves market/capital interests over social welfare
Editorial
-0.10
Structural
-0.08
SETL
-0.04
Combined
ND
Context Modifier
ND
AI product positioned for commercial capability, not human welfare or social security. No observable commitment to using AI for equitable resource distribution or social protection systems.
-0.11
Article 23Work & Equal Pay
Medium F: No discussion of labor rights implications or worker displacement P: Capability framing suggests automation of work without protection of workers
Editorial
-0.12
Structural
-0.10
SETL
-0.05
Combined
ND
Context Modifier
ND
Product emphasized for 'complex tasks' without addressing potential labor market disruption, worker training, job displacement, or fair compensation for labor automated. Missing labor protection framework.
-0.07
Article 24Rest & Leisure
Low F: No discussion of rest/leisure rights in context of AI-driven productivity pressure P: No observable work-hour protections in AI deployment
Editorial
-0.08
Structural
-0.06
SETL
-0.04
Combined
ND
Context Modifier
ND
AI positioned to increase task capacity without discussing impact on work-life balance, rest rights, or freedom from overwork. Missing protections for human leisure/recovery time.
+0.06
Article 25Standard of Living
Medium P: Accessible page design supports some users' right to adequate standard of living F: No discussion of AI healthcare/nutrition/housing applications equitably
Editorial
-0.10
Structural
+0.08
SETL
-0.13
Combined
ND
Context Modifier
ND
Structural accessibility enables participation, supporting broader inclusion. However, content does not address healthcare, housing, or nutrition applications of AI, or equity in access to such beneficial applications.
-0.11
Article 26Education
Medium F: No commitment to equitable education access via AI A: Implicit positioning of AI as luxury/advanced-capability tool, not educational public good
Editorial
-0.12
Structural
-0.10
SETL
-0.05
Combined
ND
Context Modifier
ND
AI framed for commercial advantage rather than universal education access. No discussion of using AI to promote literacy, skill development, or educational equity globally. Product access models likely restricted to paying users.
+0.09
Article 27Cultural Participation
Medium P: Blog demonstrates freedom to participate in cultural life/information sharing F: Limited discussion of how AI affects shared cultural heritage or creative participation
Editorial
+0.05
Structural
+0.03
SETL
+0.03
Combined
ND
Context Modifier
ND
Blog platform enables some cultural participation and information sharing. However, no discussion of how AI training on copyrighted works affects creators' rights, or how AI may concentrate cultural production in corporate hands.
-0.18
Article 28Social & International Order
Medium F: No discussion of international social/economic order supporting this AI's governance P: Corporate announcement model without reference to UN/global human rights frameworks
Editorial
-0.14
Structural
-0.12
SETL
-0.05
Combined
ND
Context Modifier
ND
Product development and deployment occurs within corporate governance, not international human rights order. No observable commitment to Article 28 social/economic framework. Missing acknowledgment of dependence on global human rights system.
-0.17
Article 29Duties to Community
Medium F: No commitment to duties/responsibilities in AI development P: Terms of Service do not address human rights responsibilities specific to AI
Editorial
-0.08
Structural
-0.10
SETL
+0.04
Combined
ND
Context Modifier
ND
Google's responsibilities toward community and human rights in AI development not articulated. No observable limitations on AI capability based on human rights duties. Corporate benefit positioned above community responsibilities.
-0.06
Article 30No Destruction of Rights
Medium F: No safeguards preventing AI from destroying UDHR rights P: No observable restrictions on potentially harmful AI applications
Editorial
-0.06
Structural
-0.05
SETL
-0.02
Combined
ND
Context Modifier
ND
Product announced without discussion of safeguards preventing use in rights-destructive applications (surveillance, repression, discrimination). No framework ensuring AI capability respects UDHR ceiling.