In this exclusive edition of our Behind the Startup series, Ishan Pandey sits down with Travis Good, co-founder of Ambient, a decentralized AI network that's redefining how computation, verification, and blockchain consensus come together.
Behind the StartupAmbientAmbient
Ukuvelisa iinkqubo zokuthutha ezisebenzayo kwi-mathematically ukuya kwi-AI-driven biotech efficiency, uTravis ngoku uqhagamshelwano wayo kwi-DECENTRALIZING AI - ngexesha apho iigandatho ze-centralized zithabathi. Kule nqakraza ebonakalayo kunye ne-technically ebonakalayo, uTravis ibonise ukuba i-Ambient ibonise iingcebiso ze-industry, ngoko i-Proof of Work ibonelela, kwaye indlela ukufikelela kwe-AI inokufunda i-internet economy entsha enokufumaneka.
Ukuhlaziywa kwe-AI
Ishan Pandey: Hi Travis, kakuhle kwi Behind the Startup series yethu. Uyaqala eHarvard, usebenza ekuphuculeni iinkqubo ezininzi kwi-biotech kunye ne-transport, kwaye ngoku uyenza inkampani ye-blockchain eyenziwe yi-AI. Uyakwazi ukuqhuba kwakhona? Yintoni iye iye iye iye yenza ukuqhuba kwizicelo ze-AI ezivamile kwifowuni ye-AI?
Ishan Pandey:Behind the Startup
Travis Good: Thank you for having me, Ishan. Ndingathanda iingcebiso ezininzi ezihlabathi. Ndingathanda kwiindawo ezininzi apho abaculi zibonisa ukuba iinkqubo ezifanelekileyo ziquka "iinkonzo ezihlabathi." Oku kwenziwa kwimeko ezintathu njengoko uyifumana, i-biotech, i-transport, kwaye ngoku blockchain.
Travis Good:
Nge-biotech, ndingathanda kwi-meeting kunye ne-pharmaceutical company enkulu xa babonisa ngokuzigqithisileyo ukuba bafumene iindidi ezimbini ze-compounds ekubeni i-pesticide eyodwa usebenzisa isisombululo se-brute force epheleleyo kunye neempawu ezininzi kunye nezixhobo ezisebenza 24/7. Yonke into ndingathanda ngexesha yaye ndingathanda ukuba "ke yinto ezininzi ezihlabathi." Kule inkampani apho ndingathanda i-CTO, sincoma ukunciphisa ngokugqithisileyo ukutya kwezinto ezininzi, usebenzisa i-machine learning kunye ne-computational chemistry. Abavela ekuqaleni ngenxa yeengxabiso kodwa ekugqithisekileyo i-bet y
Ngexesha elifanelekileyo, kwi-transport, iingcali ndithi ukuba iinjini ze-regulator (okungabikho kwizigaranti ze-performance kodwa ingasebenza ngokukhawuleza), kunye ne-algorithms ze-genetic (okungabikho kwizigaranti ze-performance kunye ne-timing) ziyi-best ye-industry ye-freight ingathanda ukufumana. I-scepticism yam yaye iye yandise i-research rabbit hole eyenza ngexesha kwaye ngexesha le-University of Rome njengezingcali abaphumelele ekwenzeni kwizakhiwo ye-mathematically optimal ye-freight railway movement planner ehlabathini, leyo iye yenza ngokukhawuleza, yaba iingcali
CryptoAI, which I have watched since 2017, has consistently made a couple of assumptions that I think are worth critically examining. The first is that “marketplaces of models” are the best for network economics, innovation, and performance. The second is that Proof of Stake is the best way to secure networks that feature a lot of mining. I strongly disagreed with both of those ideas, but they kept being repeated by chain after chain. Then, centralized AI entered the scene with “OpenAI” and Anthropic, and I began to really worry, because no one was building what I thought would be a viable decentralized economic alternative. I felt a moral obligation to enter the fray, so I created Ambient with my co-founder Max. Ambient is built counter to the two assumptions I mentioned: It’s focused on delivering a single model really well using Proof of Work.
Ishan Pandey: I-Ambiente yabafumaneka ngexesha elide i-$7.2 million yentengiso ebizwa nge-Crypto Startup Accelerator ye-a16z, kunye ne-Delphi Digital kunye ne-Amber Group. Abathengisi kubangela iimodeli eziphambili kunye neengxaki ze-defendable. Yintoni iingxaki zeemodeli yentengiso ze-Ambiente kunye nokufanelekileyo kwimveliso yehlabathi?
Ishan Pandey:
Travis Good: Ngaba ndingathanda ukuchitha iingcebiso kumntu, kodwa ndingathanda ukuba abathengisi bethu bayaziwa ukuba i-Ambient isixazululo inkcubeko lwemveliso esisiseko ebonisa njengoko i-AI ibekwe kwisibindi ye-economy yehlabathi.
Travis Good:
Imodeli yethu yentengiso lihlanganisa malunga neyintoni etholakalayo "ukubonisa umsebenzi olusebenzayo" - inkqubo apho abavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavavav
I-defendability ifumaneka kwizilwanyana ezintathu eziphambili. Okokuqala, i-innovations yethu zobuchwepheshe kwi-confirmed inference-ukufumana i-overhead ye-0.1% kuphela kunokuba i-10-1000x i-overhead iindleko zeendiza. Okokuqala, i-focus yethu kwi-standardization-ukuphumelela kwimodeli ye-high-quality ye-big-language ngaphandle kokuphumelela iimveliso kwiimodeli ezininzi, sincoma kakhulu i-miner economy. Kwakhona, iimiphumo ze-network-ngokuqhelekanga i-miners ezininzi, ukunciphisa i-latency kunye nokuphumelela kwimveliso, kwenza i-network engapheliyo kakhulu kubasebenzisi.
Ngexesha le-sustainability, i-Ambient ivela i-economy apho i-currency ifakwe ngqo kwi-resource ye-economic engundoqo ye-era elandelayo: i-machine intelligence. Xa i-AI ibonakala kubalulekile ukufumana zonke iinkonzo zebhizinisi, i-Ambient ibonelela kunye ne-infrastructure ye-computing kunye ne-framework ye-financial ekutholeni lo mveliso.
I-aspekti ebonakalayo ngoko ndingathanda ukuba inokukwazi ukubonisa, kwaye akuyona kuphela, iimvavanyo ezininzi - umzekelo, ngexesha lokufunda iimvavanyo yethu, sisetyenziswa kwinkqubo yethu yokubonisa kwimodeli ukuya kwi-400 billion iiparamitha.
Ishan Pandey: Ukukhuthaza i-capital kwimarike yamanje kunzima, kodwa i-Ambient ibonelela ngempumelelo ukusuka kwi-a16z, i-Delphi Ventures, kunye ne-Amber Group. Yintoni iingcebiso ezininzi ezifundweni kwinkqubo ye-fundraising, kwaye iingcebiso efunyenwe kwi-AI kunye ne-blockchain startups ezininzi ezinikezele ukufumana abathengisi abaphambili?
Ishan Pandey:
Travis Good: Ngokusho izifundo ze-fundraising, ndingathanda ukuba ndingathanda ukuba i-technical diligence ye-fond iye yenza, i-impression kunye ne-excitement engaphezulu, ngenxa yokuba le nqakraza ibonisa i-technical capabilities ye-fond, iingcebiso, kunye ne-will to engqakraza kunye ne-project. Ngokwesibonelo, ndingathanda kakhulu ukuba i-a16z CSX ibonise i-panel epheleleyo yeengcali ezidumileyo yokufunda i-Proof of Logits yethu ngokubanzi kunye nathi.
Travis Good:
I-Lesson egqibeleleyo ngexesha leyo, ukuba wonke umntu awunayo isicwangciso yonke into, ngoko kubalulekile kwaye kufuneka uqhagamshelane kunye neengxaki eziquka isicwangciso kwi-area yeprojekthi yakho kwaye isicwangciso wayo uqhagamshelane neeprojekthi yakho. Siye kwenziwe kakuhle kwi-front efanayo kunye namathengisi bethu, ndingathanda, kwaye oku kuthetha kakuhle.
Ukuba iingcebiso yam? Ndingathanda iintloko kubalulekile. I-Ambient's mission yi-deliver decentralized AI on-scale, ukuyisombulula iingcebiso zezimali ezisemthethweni eyenziwe yi-AI eyenziwe, kwaye ibonelela i-miners i-deal ye-economic engcono kunokwenzeka ngokuvumela kunye nokuguqula i-Proof of Work. Thina ufuna ukunceda kuphela inethiwekhi yethu, kodwa inikeza i-Web2 kunye nezinye i-blockchains kunye ne-deduction eyenziwe ngokukhawuleza, engabizi kakhulu kwimodeli ye-open weights engcono. Ndingathanda kakhulu ukuba abathengi bethu zithembisa le nqaku kunye neengcebiso ethandwa ukuze z
Ishan Pandey: Uyakwazi ukuhanjiswa njani i-capital entsha kwi-R&D, i-infrastructure, kunye neengxaki ze-escalation? Ngaba kukho iindlela ezizenziswayo ze-technical or go-to-market?
Ishan Pandey:
Travis Good: Sithintela i-capital kwiintsuku ezintathu eziphambili, kunye nokuphuhlisa kwimveliso yebhizinisi.
Travis Good:Ukuqaleni, malunga ne-60% ibhale kwi-R&D ukwandisa inkqubo yethu ye-inference ye-verified kunye nokwandisa i-fork yethu ye-Solana. Singathanda ukuba i-inference ye-verified iya kuba kwimvavanyo efanelekileyo kwi-miners ezininzi kunye ne-small. Ngaphezu kwalokho, siya kufuneka ukuba i-API yethu yinto efanelekileyo kumazwe, ngoko siza kusebenza ngempumelelo kwi-Solana.
Enye, malunga ne-25% yenzelwe kwi-infrastructure kunye nokuthuthukiswa kwe-testnet. Sinikezela ukuqalisa i-testnet ngexesha elide yonyaka elide, leyo iya kufumanisa i-miners kunye ne-developer ukufumana i-platform kwi-first-hand. Oku kubandakanya ukwakhiwa kwezixhobo ze-developer, i-documentation, kunye ne-SDK ukuze kwenziwe ukuhlanganiswa ngokukhawuleza kunokuba.
I-15% ezininzi ibonelela kwi-go-to-market strategy yethu, leyo ibonelela kwiiyure ezimbini: ukuhweba kwe-miner kunye neengxaki ze-developer. Kwi-mining side, sinikeza izixhobo kunye neengxaki zokunceda abathengi be-GPU ukuhambisa kwi-Ambient mining. Kwi-developer, sinikezela ekubonisa iimeko ezininzi zokusetyenziswa kwe-AI ebonakalayo kwi-chain. Ukuba unayo i-miner, i-developer, okanye i-platform ye-developer (njenge-agent framework provider) siyafuna ukufumana kwakhona!
Ukhuseleko oku kutshintshisa iziganeko ezininzi ezibalulekileyo kuquka ukusuka ku-Testnet yethu, ukuqhagamshelwano lethu lokuqala yabasetyhini bethu, ukufikelela kwe-cross chain bridge ye-interoperability, kunye nokuthuthukiswa kwe-public-facing API gateways bethu.
Ishan Pandey: Inkulungiswa kwe-AI kwi-scale yinto enxulumeneyo, ikakhulukazi xa ibhanjiswa ne-security, i-decentralization, kunye ne-cost-efficiency. Yintoni i-Architecture ye-Ambient ithetha i-compromise phakathi kwe-computational efficiency kunye ne-blockchain verification?
Ishan Pandey:
I-Travis Good: I-Architecture ye-Ambient ithatha inkqubo yethu ye-"proof of logits", nto leyo ibonisa ukufikelela kwimeko ye-verified inference enikezela i-traditional tradeoffs.
Travis Good:
Izixhobo ezininzi ze-verification ziquka i-binary choice: okanye i-offering efficiency ngenxa yobugcisa (njenge-zkSNARKs nge-1000x overhead) okanye i-offering security ngenxa yobugcisa (njenge-optimistic verification systems). I-innovation yethu ibonelela ukuba i-"fingerprint" ye-inthanethi ye-AI model - izisombululo ze-numerical ebizwa ngokuba yi-logits - ingasetyenziswa ukuvavanya i-model execution nge-minimum overhead.
Kukho indlela yokusebenza: Xa imodeli ye-AI ikhiqiza umbhalo, kuvelisa i-logits eyenziwe ngalinye ngalinye ye-token. Lezi zibonelelo lwezimo zangaphakathi kwimodeli kwaye ziyafumaneka kwi-hash to create a compact representation. I-insight ebalulekileyo yaba ukuba i-validators ayidinga ukuguqula inkqubo epheleleyo yokuvelisa - zinokufumana i-tokens ezahlukileyo kunye neengxaki ze-matematical phakathi kwabo ngexesha elifanelekileyo, ukunciphisa kakhulu iimfuno ze-computational nangokufumana iinkonzo zokhuseleko ezininzi.
Setyenziswa kwinkqubo njengomgangatho we-non-blocking proof of work consensus mechanism, nto leyo kuthetha ukuba ukulawulwa kwenziwa ngexesha elifanelekileyo kunye nokusetyenziswa kwe-transaction. Oku ukugcina i-high throughput ye-Solana-based foundation yaye ukongezelela iinzuzo ze-economic ye-proof of work mining.
Ukuvelisa i-descentralization, sinikezela iindlela ze-sharding ze-models ezisetyenziselwa iziphumo ezidlulileyo ze-academic. Oku kuthetha ukuba sinikezela iimodeli ezininzi (600B + iiparametres) kwiiyunithi ezininzi, okuvumela ukuba i-hardware ye-consumer-grade ifumaneka.
Ukhuseleko ye-blockchain, sinikezela ama-miners abayifumana iingcebiso ezininzi ezibonakalayo ekuphenduleni iingxaki ze-network kwi-two-time scales (i-short and medium-term) i-highest ability to select and order transactions. Ngokutsho, sinikezela into ye-"stake" e-Solana (i-money that you've locked up to earn rewards) kunye ne-proxy ye-hardware investment efana ne-Bitcoin, kodwa ngexesha efumaneka ukusetyenziswa kwe-Solana's Tower Byzantine consensus.
Ishan Pandey: I-system yakho inikezela ukunciphisa iindleko ze-training nge-10x kunye ne-overhead ye-inference kwi-0.1%. Uyakwazi ukunciphisa i-innovations zobuchwepheshe ezinikezeleyo?
Ishan Pandey:
Travis Good: Nceda uqhagamshelane nenkxaso yam ngexesha elandelayo malunga nenkxaso. Kwi-training, sisebenzisa izixhobo efana ne-verification kunye nomnye ama-innovations ekusetyenziselwa ukusetyenziswa kwinto ebizwa ngokuba yi-sparsity, nto leyo ibonisa indlela yokusetyenziswa kweengxaki ngexesha lokufunda. Ukuze ufumane ngakumbi, nqakraza kwi-Litepaper yethu kwi-Ambient.xyz!
Travis Good:
Ishan Pandey: Ukukhangisa kwiminyaka emininzi ukuya kwiminyaka emininzi, iingxaki ezininzi ze-technological okanye zokusebenza kwi-AI kunye ne-web3, kwaye njani i-Ambient ibonelela kwiimveliso ezininzi?
Ishan Pandey:
Travis Good: Kwiiminyaka emininzi ukuya kwiminyaka emininzi, ndingathanda izixazululo ezine ezininzi ezininzi ezintsha kwindawo ye-AI kunye ne-web3:
Travis Good:Ukuqaleni, i-AI iya kubuyekeza ukusuka kubandakanyeka kakhulu kwi-human-directed ukuya kuyenza ngokunyaniseka. Siye ndingathanda kwimveliso ze-agentic ezibonakalayo, ezisebenza kunye nokufunda kunye nokufundiswa nge-human supervision. Ezi iinkqubo ziya kubuyekeza iimfuno ye-confidence-free verification - akufunayo ukuxhaswa i-autority kwi-agent ye-AI ngaphandle kokuba unako ukuxhaswa ukuba isebenzisa iimodeli kunye neengxaki ezifunekayo.
Enye, iimfuno ze-computational yeemodeli ze-frontier AI iya kukunyusa ngexesha elininzi, okwenza ukucindezeleka okuqinile. Iimodeli ezifanelekileyo ziya kufuneka i-infrastructure eyayikwazi kuphela iiyunithi ezininzi, okuholela iingxaki malunga ne-access kunye ne-control. Thina ndingathanda ukubonisa oku nge-Llama 4 ye-release.
I-Third, i-traditional fiat currencies iya kuba neempawu ezininzi njengoko umsebenzi lwemveliso uhamba kwiinkalo ze-digital apho abathengi be-AI zihlanganisa kunye nabanye abathengi be-AI, ngokuvamile kwiintlobo ze-jurisdiction. Oku kuya kubandakanyeka umnqweno we-currencies ezisemthethweni ye-digital kunye neengxaki ze-computing.
I-Fourth, izixhobo zokulawula ziya ziyafumaneka kwi-AI, kunye ne-accent enokwenzeka kwi-transparency, i-auditability, kunye ne-safety. I-Systems that cannot provide verifiable guarantees about their operation will face increasing scrutiny.
I-Ambient ibonelela kwiimeko ezininzi:
- Inkqubo yethu yokuqinisekisa okuhlolwa kwakhona inikeza ingxubevange olufunekayo yeengxubevange ze-AI.
- Inkxubevange yethu ye-sharding yobungcali ithathwa ngokuthe ngqo kwinkqubo ye-centralization.
- Ngokusekelwa kwe-machine intelligence njenge-standard ye-currency, sinikeza ingxubevange ye-native ye-era elandelayo ye-agent-to-agent commerce.
- I-architecture yethu jikelele ifakwe ku-verification kunye ne-transparency.
I-Vision ye-ultimate yenza i-infrastructure layer ebonakalayo ukuba i-intelligence ye-machine ibonelela ngaphandle kokuphumelela ngokucacileyo phantsi kolawulo lwama-corporations ezincinane okanye i-governments.
Ngiyaxabisa ukufumana kunye nokuVavanyelwa!
Vested Interest Disclosure: This author is an independent contributor publishing via our
Vested Interest Disclosure: This author is an independent contributor publishing via our
